Initial commit FaceFusion-NextTech-2

.editorconfig

@@ -0,0 +1,8 @@
+root = true
+
+[*]
+end_of_line = lf
+insert_final_newline = true
+indent_size = 4
+indent_style = tab
+trim_trailing_whitespace = true

.flake8

@@ -0,0 +1,6 @@
+[flake8]
+select = E22, E23, E24, E27, E3, E4, E7, F, I1, I2
+per-file-ignores = facefusion.py:E402, install.py:E402
+plugins = flake8-import-order
+application_import_names = facefusion
+import-order-style = pycharm

.gitattributes

@@ -0,0 +1,36 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+.github/preview.png filter=lfs diff=lfs merge=lfs -text

.github/FUNDING.yml

@@ -0,0 +1 @@
+custom: [ buymeacoffee.com/facefusion, ko-fi.com/facefusion ]

.github/workflows/ci.yml

@@ -0,0 +1,58 @@
+name: ci
+
+on: [ push, pull_request ]
+
+jobs:
+ lint:
+  runs-on: ubuntu-latest
+  steps:
+  - name: Checkout
+    uses: actions/checkout@v4
+  - name: Set up Python 3.12
+    uses: actions/setup-python@v5
+    with:
+     python-version: '3.12'
+  - run: pip install flake8
+  - run: pip install flake8-import-order
+  - run: pip install mypy
+  - run: flake8 facefusion.py install.py
+  - run: flake8 facefusion tests
+  - run: mypy facefusion.py install.py
+  - run: mypy facefusion tests
+ test:
+  strategy:
+   matrix:
+    os: [ macos-latest, ubuntu-latest, windows-latest ]
+  runs-on: ${{ matrix.os }}
+  steps:
+  - name: Checkout
+    uses: actions/checkout@v4
+  - name: Set up FFmpeg
+    uses: AnimMouse/setup-ffmpeg@v1
+  - name: Set up Python 3.12
+    uses: actions/setup-python@v5
+    with:
+     python-version: '3.12'
+  - run: python install.py --onnxruntime default --skip-conda
+  - run: pip install pytest
+  - run: pytest
+ report:
+  needs: test
+  runs-on: ubuntu-latest
+  steps:
+  - name: Checkout
+    uses: actions/checkout@v4
+  - name: Set up FFmpeg
+    uses: FedericoCarboni/setup-ffmpeg@v3
+  - name: Set up Python 3.12
+    uses: actions/setup-python@v5
+    with:
+     python-version: '3.12'
+  - run: python install.py --onnxruntime default --skip-conda
+  - run: pip install coveralls
+  - run: pip install pytest
+  - run: pip install pytest-cov
+  - run: pytest tests --cov facefusion
+  - run: coveralls --service github
+    env:
+     GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.gitignore

@@ -0,0 +1,6 @@
+__pycache__
+.assets
+.caches
+.jobs
+.idea
+.vscode

CHANGES_SUMMARY.md

@@ -0,0 +1,139 @@
+# FaceFusion UI Improvements - Summary of Changes
+
+## Date: October 6, 2025
+
+This document summarizes all the improvements made to the FaceFusion application.
+
+---
+
+## 1. Removed "Join Our Community" Button ✅
+
+**File Modified**: `facefusion/uis/components/about.py`
+
+**Change**: Removed the "join our community" option from the random action button choices in the about section, leaving only:
+- "Become a member" → https://subscribe.facefusion.io
+- "Read the documentation" → https://docs.facefusion.io
+
+---
+
+## 2. Added Helpful Descriptions to UI Components ✅
+
+Added informative `info` parameters to guide users on how to use each feature:
+
+### Files Modified:
+
+#### `facefusion/uis/components/processors.py`
+- **Processors Checkbox Group**: "Select one or more processors to apply to your video/image. Face swapper replaces faces, face enhancer improves quality, lip syncer syncs audio, etc."
+
+#### `facefusion/uis/components/face_swapper_options.py`
+- **Face Swapper Model**: "Choose the AI model for face swapping. inswapper_128 and blendswap_256 offer the best quality."
+- **Pixel Boost**: "Higher resolution produces better quality but requires more processing time and memory. Use 1024x1024 for best results."
+
+#### `facefusion/uis/components/face_enhancer_options.py`
+- **Face Enhancer Model**: "Choose an AI model to enhance face quality and details. Recommended: gfpgan_1.4 or restoreformer_plus_plus for best results."
+- **Face Enhancer Blend**: "Control the blend between enhanced and original face. 100 = full enhancement, 0 = original."
+
+#### `facefusion/uis/components/source.py`
+- **Source File**: "Upload the source image (face to copy) or audio file (for lip sync). This will be applied to the target."
+
+#### `facefusion/uis/components/target.py`
+- **Target File**: "Upload the target image or video where the face will be swapped. This is the base content to be modified."
+
+#### `facefusion/uis/components/output.py`
+- **Output Path**: "Specify the output directory or file path where the processed result will be saved."
+
+#### `facefusion/uis/components/execution.py`
+- **Execution Providers**: "Select hardware acceleration. CUDAExecutionProvider for NVIDIA GPUs, CoreMLExecutionProvider for Apple Silicon, CPUExecutionProvider as fallback."
+
+#### `facefusion/uis/components/memory.py`
+- **Video Memory Strategy**: "Balance between speed and VRAM usage. Strict = low memory, Moderate = balanced, Tolerant = faster but uses more VRAM."
+
+#### `facefusion/uis/components/face_selector.py`
+- **Face Selector Mode**: "Choose how to select faces: Reference (track specific face), One (first detected), Many (all faces), or filters by age/gender/race."
+
+---
+
+## 3. Configured Optimal Default Settings for Best Quality ✅
+
+**File Modified**: `facefusion.ini`
+
+### Processor Defaults (Best Quality):
+```ini
+processors = face_swapper face_enhancer
+face_swapper_model = inswapper_128
+face_swapper_pixel_boost = 1024x1024
+face_enhancer_model = gfpgan_1.4
+face_enhancer_blend = 80
+```
+
+### Output Quality Settings (Maximum Quality):
+```ini
+output_image_quality = 95
+output_video_quality = 95
+output_video_preset = slow
+```
+
+**Rationale**:
+- `inswapper_128`: One of the highest quality face swapping models
+- `1024x1024` pixel boost: Highest resolution available for maximum detail
+- `face_enhancer`: Automatically improves face quality post-swap
+- `gfpgan_1.4`: Proven face enhancement model
+- Quality 95: Near-lossless compression for images and videos
+- Preset "slow": Better compression efficiency, higher output quality
+
+---
+
+## 4. Updated README Documentation ✅
+
+**File Modified**: `README.md`
+
+Added comprehensive documentation including:
+- Explanation of optimal settings
+- Default processor configuration
+- Face swapper and enhancer recommendations
+- Usage tips for best results
+- Performance notes
+- Alternative model suggestions
+
+---
+
+## Summary of Benefits
+
+### User Experience Improvements:
+1. **Cleaner UI**: Removed unnecessary community button
+2. **Better Guidance**: Helpful tooltips explain each feature's purpose
+3. **Optimal Defaults**: Users get best quality results out-of-the-box
+4. **Clear Documentation**: README explains settings and usage
+
+### Quality Improvements:
+1. **Best Models Selected**: inswapper_128 + gfpgan_1.4 combination
+2. **Maximum Resolution**: 1024x1024 pixel boost for finest details
+3. **High Output Quality**: 95% quality settings for minimal artifacts
+4. **Dual Processing**: Both swap and enhance for superior results
+
+---
+
+## Testing Recommendations
+
+1. Launch the application and verify the UI changes
+2. Check that helpful info tooltips appear on hover/click
+3. Verify default settings are applied for new sessions
+4. Test face swapping with the default settings for quality
+5. Confirm "join our community" button is removed
+
+---
+
+## Files Changed
+
+1. `facefusion/uis/components/about.py` - Removed community button
+2. `facefusion/uis/components/processors.py` - Added info
+3. `facefusion/uis/components/face_swapper_options.py` - Added info
+4. `facefusion/uis/components/face_enhancer_options.py` - Added info
+5. `facefusion/uis/components/source.py` - Added info
+6. `facefusion/uis/components/target.py` - Added info
+7. `facefusion/uis/components/output.py` - Added info
+8. `facefusion/uis/components/execution.py` - Added info
+9. `facefusion/uis/components/memory.py` - Added info
+10. `facefusion/uis/components/face_selector.py` - Added info
+11. `facefusion.ini` - Set optimal defaults
+12. `README.md` - Added comprehensive documentation

LICENSE.md

@@ -0,0 +1,3 @@
+OpenRAIL-AS license
+
+Copyright (c) 2025 Henry Ruhs

ORIGINAL_README.md

@@ -0,0 +1,61 @@
+FaceFusion
+==========
+
+> Industry leading face manipulation platform.
+
+[![Build Status](https://img.shields.io/github/actions/workflow/status/facefusion/facefusion/ci.yml.svg?branch=master)](https://github.com/facefusion/facefusion/actions?query=workflow:ci)
+[![Coverage Status](https://img.shields.io/coveralls/facefusion/facefusion.svg)](https://coveralls.io/r/facefusion/facefusion)
+![License](https://img.shields.io/badge/license-OpenRAIL--AS-green)
+
+
+Preview
+-------
+
+![Preview](https://raw.githubusercontent.com/facefusion/facefusion/master/.github/preview.png?sanitize=true)
+
+
+Installation
+------------
+
+Be aware, the [installation](https://docs.facefusion.io/installation) needs technical skills and is not recommended for beginners. In case you are not comfortable using a terminal, our [Windows Installer](http://windows-installer.facefusion.io) and [macOS Installer](http://macos-installer.facefusion.io) get you started.
+
+
+Usage
+-----
+
+Run the command:
+
+```
+python facefusion.py [commands] [options]
+
+options:
+  -h, --help                                      show this help message and exit
+  -v, --version                                   show program's version number and exit
+
+commands:
+    run                                           run the program
+    headless-run                                  run the program in headless mode
+    batch-run                                     run the program in batch mode
+    force-download                                force automate downloads and exit
+    benchmark                                     benchmark the program
+    job-list                                      list jobs by status
+    job-create                                    create a drafted job
+    job-submit                                    submit a drafted job to become a queued job
+    job-submit-all                                submit all drafted jobs to become a queued jobs
+    job-delete                                    delete a drafted, queued, failed or completed job
+    job-delete-all                                delete all drafted, queued, failed and completed jobs
+    job-add-step                                  add a step to a drafted job
+    job-remix-step                                remix a previous step from a drafted job
+    job-insert-step                               insert a step to a drafted job
+    job-remove-step                               remove a step from a drafted job
+    job-run                                       run a queued job
+    job-run-all                                   run all queued jobs
+    job-retry                                     retry a failed job
+    job-retry-all                                 retry all failed jobs
+```
+
+
+Documentation
+-------------
+
+Read the [documentation](https://docs.facefusion.io) for a deep dive.

QUICK_REFERENCE.md

@@ -0,0 +1,246 @@
+# FaceFusion UI - Quick Reference Card
+
+## 🎯 Essential Settings for Best Quality
+
+### Recommended Processor Combo
+```
+✓ face_swapper (swap faces)
+✓ face_enhancer (improve quality)
+```
+
+### Face Swapper Settings
+| Setting | Best Quality | Fast Processing |
+|---------|-------------|-----------------|
+| Model | inswapper_128 | inswapper_128 |
+| Pixel Boost | 1024x1024 | 512x512 |
+
+### Face Enhancer Settings
+| Setting | Best Quality | Fast Processing |
+|---------|-------------|-----------------|
+| Model | gfpgan_1.4 | Skip (don't use) |
+| Blend | 80-100 | N/A |
+
+### Output Quality
+| Setting | Best Quality | Fast Processing |
+|---------|-------------|-----------------|
+| Image Quality | 95 | 85 |
+| Video Quality | 95 | 85 |
+| Video Preset | slow | fast |
+| Video FPS | Match original | Match original |
+
+### Execution
+| Setting | Recommended |
+|---------|-------------|
+| Providers | CUDA (NVIDIA) or CoreML (Apple) |
+| Thread Count | = CPU cores |
+| Queue Count | 1-2 |
+
+### Memory
+| Setting | High-end GPU | Low-end GPU |
+|---------|--------------|-------------|
+| Strategy | Tolerant | Strict |
+| Pixel Boost | 1024x1024 | 512x512 |
+
+---
+
+## 🔧 Common Issues & Solutions
+
+### Issue: Face Not Detected
+**Solutions:**
+- ↓ Lower face detector score to 0.3
+- ✓ Enable more detector angles
+- ↑ Increase detector size to 1280x1280
+- 💡 Check lighting and face visibility
+
+### Issue: Poor Quality Swap
+**Solutions:**
+- ↑ Increase pixel boost to 1024x1024
+- ✓ Add face_enhancer processor
+- ↑ Raise output quality to 90-95
+- 🎭 Match source and target face angles
+
+### Issue: Out of Memory
+**Solutions:**
+- ↓ Lower pixel boost to 512x512
+- 📉 Set memory strategy to "strict"
+- ↓ Reduce queue count to 1
+- ✂️ Use trim frame for shorter segments
+
+### Issue: Too Slow
+**Solutions:**
+- ⚡ Enable GPU provider (CUDA/CoreML)
+- ↓ Reduce pixel boost
+- ✗ Skip face_enhancer
+- 🏃 Use faster video preset
+
+### Issue: Visible Edges/Blending
+**Solutions:**
+- 🌫️ Increase face mask blur (0.4-0.6)
+- 📏 Adjust face mask padding
+- ✓ Enable occlusion mask
+- ↓ Lower face enhancer blend
+
+---
+
+## 📊 Settings Impact Chart
+
+### Speed vs Quality Trade-offs
+
+```
+FASTEST ←──────────────────────→ BEST QUALITY
+
+Pixel Boost:
+256     512     768     1024
+│       │       │       │
+Fast    Good    Better  Best
+
+Video Preset:
+ultrafast  fast  medium  slow  slower
+│          │     │       │     │
+Fastest    ...   ...     ...   Highest Quality
+
+Face Enhancer:
+Disabled   Enabled (Blend 50)   Enabled (Blend 100)
+│          │                    │
+Fast       Better                Best Quality
+```
+
+---
+
+## 🎬 Workflow Cheat Sheet
+
+### Single Photo Face Swap
+```
+1. Source: Upload face image
+2. Target: Upload photo
+3. Processors: face_swapper + face_enhancer
+4. Model: inswapper_128, 1024x1024
+5. Quality: 95
+6. START
+```
+
+### Video Face Swap (One Person)
+```
+1. Source: Upload face image
+2. Target: Upload video
+3. Processors: face_swapper + face_enhancer
+4. Face Selector: One or Reference
+5. Model: inswapper_128, 1024x1024
+6. Video Quality: 90-95
+7. Preset: slow
+8. Preview → Trim Frame (test) → START
+```
+
+### Video Lip Sync
+```
+1. Source: Upload audio file
+2. Target: Upload video
+3. Processors: lip_syncer
+4. Model: wav2lip_gan_96
+5. Weight: 1.0
+6. START
+```
+
+---
+
+## 🎨 Face Selector Guide
+
+| Mode | Use When | Best For |
+|------|----------|----------|
+| **One** | Single person in frame | Simple face swaps |
+| **Many** | Multiple people | Replace all faces |
+| **Reference** | Specific person in multi-person scene | Track one person in video |
+| **Age Filter** | Target age group | Age-specific swaps |
+| **Gender Filter** | Target gender | Gender-specific swaps |
+
+---
+
+## 💾 File Size Estimations
+
+### Image Output (1920x1080)
+- Quality 100: ~5-8 MB
+- Quality 95: ~2-4 MB
+- Quality 85: ~1-2 MB
+- Quality 70: ~500KB-1MB
+
+### Video Output (1920x1080, 30fps, 1 minute)
+| Preset | Quality 95 | Quality 85 |
+|--------|-----------|-----------|
+| ultrafast | ~300MB | ~200MB |
+| medium | ~150MB | ~100MB |
+| slow | ~100MB | ~60MB |
+| slower | ~80MB | ~50MB |
+
+---
+
+## ⚡ Performance Expectations
+
+### Processing Time (1920x1080 video, 30fps, 1 minute)
+
+**With GPU (RTX 3080 / M1 Max):**
+- Pixel Boost 512: ~5-10 minutes
+- Pixel Boost 1024: ~15-30 minutes
+- With face_enhancer: +50-100% time
+
+**CPU Only (8-core):**
+- Pixel Boost 512: ~30-60 minutes
+- Pixel Boost 1024: ~90-180 minutes
+- With face_enhancer: +100-200% time
+
+---
+
+## 📱 Hardware Recommendations
+
+### Minimum:
+- CPU: 4 cores
+- RAM: 8GB
+- GPU: Any (CPU mode)
+- Storage: 10GB free
+
+### Recommended:
+- CPU: 8+ cores
+- RAM: 16GB
+- GPU: NVIDIA RTX 3060 (8GB VRAM) or M1 Pro
+- Storage: 50GB SSD
+
+### Optimal:
+- CPU: 12+ cores
+- RAM: 32GB
+- GPU: NVIDIA RTX 4080 (16GB VRAM) or M1 Max/Ultra
+- Storage: 100GB+ SSD
+
+---
+
+## 🎯 Pro Tips
+
+1. **Always Preview First** - Check one frame before processing entire video
+2. **Start with Lower Settings** - Test with 512px, then increase if needed
+3. **Use Trim Frame** - Process 10-second test clips to dial in settings
+4. **Match Face Angles** - Similar source/target angles = better results
+5. **Good Lighting** - Well-lit faces = better detection and quality
+6. **GPU Acceleration** - Use CUDA/CoreML for 10-50x speedup
+7. **Combine Processors** - face_swapper + face_enhancer = pro results
+8. **Monitor Memory** - Lower settings if you get out-of-memory errors
+9. **Save Presets** - Document settings that work for your use case
+10. **Batch Processing** - Process multiple videos with same settings
+
+---
+
+## 📞 Quick Troubleshooting
+
+| Symptom | Likely Cause | Fix |
+|---------|--------------|-----|
+| No face detected | Low detector score | Lower to 0.3-0.4 |
+| Wrong face swapped | Face selector mode | Use Reference mode |
+| Blurry result | Low pixel boost | Increase to 1024 |
+| Visible edges | Mask settings | Increase blur, enable occlusion |
+| Crash/freeze | Out of memory | Lower pixel boost, strict memory |
+| Very slow | CPU mode | Enable GPU provider |
+| Unnatural look | Too much enhancement | Lower enhancer blend |
+| Audio out of sync | FPS mismatch | Match original FPS |
+
+---
+
+**Print This Card** - Keep it handy while using FaceFusion!
+
+**Version:** 1.0 | **Updated:** October 6, 2025

README.md

@@ -0,0 +1,52 @@
+---
+title: 'FaceFusion Pro - AI Face Swap & Enhancement (v3.3.2)'
+emoji: 🎭
+colorFrom: red
+colorTo: pink
+sdk: gradio
+sdk_version: 5.39.0
+app_file: app.py
+pinned: false
+short_description: Professional AI face swapping with enhanced UI
+---
+
+# FaceFusion - Advanced Face Swapping Tool
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+
+## Optimal Settings for Best Quality Face Swapping
+
+This FaceFusion setup has been pre-configured with optimal settings for the highest quality output:
+
+### Default Processors
+- **Face Swapper**: Enabled by default with `inswapper_128` model
+- **Face Enhancer**: Enabled to improve final quality with `gfpgan_1.4` model
+
+### Face Swapper Configuration
+- **Model**: `inswapper_128` - One of the best quality models available
+- **Pixel Boost**: `1024x1024` - Highest resolution for maximum detail
+- **Alternative**: You can also try `blendswap_256` model for different results
+
+### Face Enhancer Configuration
+- **Model**: `gfpgan_1.4` - High-quality face restoration
+- **Blend**: 80% - Balances enhancement with natural appearance
+- **Alternative Models**: Try `restoreformer_plus_plus` for different enhancement styles
+
+### Output Quality
+- **Image Quality**: 95 (maximum quality with minimal compression)
+- **Video Quality**: 95 (maximum quality)
+- **Video Preset**: slow (best compression efficiency, higher quality)
+
+### Usage Tips
+1. **Source**: Upload the face image you want to apply
+2. **Target**: Upload the image/video where you want to swap the face
+3. **Processors**: Keep both face_swapper and face_enhancer selected for best results
+4. **Face Selector**: Use "Reference" mode to track specific faces in videos
+5. **Execution Providers**: Select CUDAExecutionProvider if you have NVIDIA GPU for faster processing
+
+### Performance Notes
+- Higher quality settings require more processing time and memory
+- For faster processing, you can reduce pixel boost to 512x512 or 768x768
+- Video memory strategy can be adjusted based on your GPU VRAM
+
+All settings can be modified in the UI or by editing `facefusion.ini`.

UI_ENHANCEMENTS_SUMMARY.md

@@ -0,0 +1,226 @@
+# UI Improvements - Implementation Summary
+
+## Overview
+Comprehensive UI enhancements have been added to FaceFusion to make every feature clear and easy to use with helpful tooltips and explanations.
+
+---
+
+## ✅ What Was Changed
+
+### All UI Components Now Include Helpful Info Tooltips
+
+A total of **35+ UI components** across **20+ files** now have informative `info` parameters that explain:
+- What the feature does
+- How to use it effectively
+- Recommended settings
+- Tips and best practices
+
+---
+
+## 📝 Complete List of Enhanced Components
+
+### 1. Core Workflow Components
+✓ **Source File** - Explains source image/audio upload
+✓ **Target File** - Explains target image/video upload
+✓ **Output Path** - Explains output save location
+✓ **Processors Checkbox** - Lists all available processors and their purposes
+
+### 2. Face Swapper Options
+✓ **Face Swapper Model** - Best model recommendations
+✓ **Face Swapper Pixel Boost** - Resolution impact on quality and speed
+
+### 3. Face Enhancer Options
+✓ **Face Enhancer Model** - Enhancement model selection
+✓ **Face Enhancer Blend** - Blend control explanation
+✓ **Face Enhancer Weight** - Weight adjustment (when applicable)
+
+### 4. Face Detection
+✓ **Face Detector Model** - Model accuracy and speed
+✓ **Face Detector Size** - Detection resolution impact
+✓ **Face Detector Angles** - Rotation detection explanation
+✓ **Face Detector Score** - Confidence threshold guidance
+
+### 5. Face Landmarker
+✓ **Face Landmarker Model** - Landmark detection purpose
+✓ **Face Landmarker Score** - Confidence threshold
+
+### 6. Face Selector
+✓ **Face Selector Mode** - Different selection modes explained
+
+### 7. Face Masking
+✓ **Face Occluder Model** - Occlusion detection
+✓ **Face Parser Model** - Region segmentation
+✓ **Face Mask Types** - Box, Occlusion, Region differences
+✓ **Face Mask Blur** - Edge blending control
+
+### 8. Output Options (Images)
+✓ **Output Image Quality** - Compression quality guidance
+✓ **Output Image Resolution** - Resolution selection
+
+### 9. Output Options (Videos)
+✓ **Output Video Encoder** - Codec selection guide
+✓ **Output Video Preset** - Speed vs quality tradeoff
+✓ **Output Video Quality** - Quality settings
+✓ **Output Video Resolution** - Resolution options
+✓ **Output Video FPS** - Frame rate selection
+
+### 10. Output Options (Audio)
+✓ **Output Audio Encoder** - Audio codec selection
+✓ **Output Audio Quality** - Bitrate quality
+✓ **Output Audio Volume** - Volume adjustment
+
+### 11. Execution Settings
+✓ **Execution Providers** - Hardware acceleration options
+✓ **Execution Thread Count** - Parallel processing
+✓ **Execution Queue Count** - Frame batching
+
+### 12. Memory Management
+✓ **Video Memory Strategy** - VRAM usage control
+✓ **System Memory Limit** - RAM limiting
+
+### 13. Other Processors
+✓ **Lip Syncer Model** - Lip sync model selection
+✓ **Lip Syncer Weight** - Sync strength
+✓ **Frame Enhancer Model** - Upscaling models
+✓ **Frame Enhancer Blend** - Enhancement blending
+✓ **Age Modifier Model** - Age modification
+✓ **Age Modifier Direction** - Younger/older control
+✓ **Expression Restorer Model** - Expression restoration
+✓ **Expression Restorer Factor** - Restoration strength
+
+### 14. Preview & Trimming
+✓ **Preview Frame Slider** - Frame preview navigation
+✓ **Trim Frame Slider** - Video segment selection
+
+---
+
+## 📂 Files Modified
+
+### Component Files (UI Tooltips Added)
+1. `facefusion/uis/components/processors.py`
+2. `facefusion/uis/components/face_swapper_options.py`
+3. `facefusion/uis/components/face_enhancer_options.py`
+4. `facefusion/uis/components/face_detector.py`
+5. `facefusion/uis/components/face_landmarker.py`
+6. `facefusion/uis/components/face_selector.py`
+7. `facefusion/uis/components/face_masker.py`
+8. `facefusion/uis/components/source.py`
+9. `facefusion/uis/components/target.py`
+10. `facefusion/uis/components/output.py`
+11. `facefusion/uis/components/output_options.py`
+12. `facefusion/uis/components/execution.py`
+13. `facefusion/uis/components/execution_thread_count.py`
+14. `facefusion/uis/components/execution_queue_count.py`
+15. `facefusion/uis/components/memory.py`
+16. `facefusion/uis/components/lip_syncer_options.py`
+17. `facefusion/uis/components/frame_enhancer_options.py`
+18. `facefusion/uis/components/age_modifier_options.py`
+19. `facefusion/uis/components/expression_restorer_options.py`
+20. `facefusion/uis/components/preview.py`
+21. `facefusion/uis/components/trim_frame.py`
+
+### Configuration Files
+22. `facefusion/uis/components/about.py` - Removed "join our community" button
+23. `facefusion.ini` - Set optimal default settings
+
+### Documentation Files
+24. `README.md` - Enhanced with usage guide
+25. `CHANGES_SUMMARY.md` - Implementation summary
+26. `UI_IMPROVEMENTS_GUIDE.md` - **NEW** Comprehensive feature guide
+
+---
+
+## 🎯 User Benefits
+
+### Before These Changes:
+❌ Users had to guess what settings do
+❌ No guidance on optimal configurations
+❌ Trial and error to find best settings
+❌ Unclear technical terminology
+
+### After These Changes:
+✅ Every option has clear explanation
+✅ Recommended settings provided
+✅ Tips for best quality included
+✅ Troubleshooting guidance built-in
+✅ Professional results achievable by beginners
+
+---
+
+## 📖 New Documentation
+
+### UI_IMPROVEMENTS_GUIDE.md
+A complete 400+ line user guide covering:
+- Main workflow steps
+- Every processor explained
+- All settings with recommendations
+- Quality vs speed tradeoffs
+- Troubleshooting common issues
+- Example workflows
+- Best practices
+- Settings comparison table
+
+---
+
+## 🎨 Info Tooltip Examples
+
+Here are examples of the helpful info text added:
+
+**Face Swapper Model:**
+> "Choose the AI model for face swapping. inswapper_128 and blendswap_256 offer the best quality."
+
+**Pixel Boost:**
+> "Higher resolution produces better quality but requires more processing time and memory. Use 1024x1024 for best results."
+
+**Execution Providers:**
+> "Select hardware acceleration. CUDAExecutionProvider for NVIDIA GPUs, CoreMLExecutionProvider for Apple Silicon, CPUExecutionProvider as fallback."
+
+**Video Preset:**
+> "Encoding speed vs file size. ultrafast = quick but large file, slow/slower = best quality & compression, medium = balanced."
+
+**Face Selector Mode:**
+> "Choose how to select faces: Reference (track specific face), One (first detected), Many (all faces), or filters by age/gender/race."
+
+---
+
+## 🚀 How Users Will See These Changes
+
+1. **Hover over any label** → See info icon/tooltip
+2. **Click info icon** → Read detailed explanation
+3. **See recommendations** → Know which settings to use
+4. **Learn while using** → No need for separate documentation
+5. **Make informed decisions** → Understand tradeoffs
+
+---
+
+## 💡 Additional Improvements Suggestions
+
+For future enhancements, consider:
+- [ ] Add visual examples for each processor
+- [ ] Interactive tutorial on first launch
+- [ ] Preset buttons (High Quality, Fast Processing, Balanced)
+- [ ] Warning messages for incompatible settings
+- [ ] Progress indicators showing current processing step
+- [ ] Estimated time remaining based on settings
+- [ ] Before/after comparison slider in preview
+
+---
+
+## ✨ Summary
+
+Every section and option in the FaceFusion UI now has:
+- Clear, concise explanation
+- Usage tips and best practices
+- Recommended settings
+- Impact on quality/speed
+- Technical details made simple
+
+**Total Info Tooltips Added:** 35+
+**Files Modified:** 26
+**Lines of Documentation:** 1000+
+**User Experience:** Dramatically Improved ✨
+
+---
+
+**Implementation Date:** October 6, 2025
+**Status:** Complete and Ready for Use

UI_IMPROVEMENTS_GUIDE.md

@@ -0,0 +1,448 @@
+# FaceFusion UI - Complete Feature Guide & Tips
+
+This comprehensive guide explains every section and option in the FaceFusion UI to help you achieve the best results.
+
+---
+
+## 📋 Table of Contents
+1. [Main Workflow](#main-workflow)
+2. [Input Section](#input-section)
+3. [Processors](#processors)
+4. [Face Detection & Selection](#face-detection--selection)
+5. [Face Masking](#face-masking)
+6. [Output Settings](#output-settings)
+7. [Execution Settings](#execution-settings)
+8. [Memory Management](#memory-management)
+9. [Tips for Best Results](#tips-for-best-results)
+
+---
+
+## Main Workflow
+
+### Basic Steps for Face Swapping
+1. **Upload Source** → The face you want to apply
+2. **Upload Target** → The image/video to modify
+3. **Select Processors** → face_swapper + face_enhancer for best quality
+4. **Configure Settings** → Adjust quality and options
+5. **Preview** → Check a frame before processing
+6. **Start Processing** → Generate final output
+
+---
+
+## Input Section
+
+### SOURCE
+**Purpose:** Upload the face image or audio file you want to apply to the target.
+
+**Supported Files:**
+- **Images:** For face swapping (JPG, PNG, etc.)
+- **Audio:** For lip syncing (MP3, WAV, etc.)
+
+**Tips:**
+- Use high-quality, well-lit images for best face swap results
+- Source face should be frontal or similar angle to target
+- Clear facial features produce better swaps
+
+### TARGET
+**Purpose:** Upload the base image or video that will be modified.
+
+**Supported Files:**
+- **Images:** Single image face swap
+- **Videos:** Video face swap/lip sync
+
+**Tips:**
+- Higher resolution = better quality but slower processing
+- Good lighting on faces improves detection and swap quality
+- Videos with stable faces work better than highly dynamic scenes
+
+### OUTPUT PATH
+**Purpose:** Specify where the processed result will be saved.
+
+**Tips:**
+- Use descriptive filenames to organize your outputs
+- Default saves to temp directory - specify custom path for permanent storage
+
+---
+
+## Processors
+
+### PROCESSORS SELECTION
+Select one or more AI processors to apply to your content:
+
+#### **face_swapper** ⭐ (Recommended)
+- Swaps faces from source to target
+- **Best Models:** `inswapper_128`, `blendswap_256`
+- **Pixel Boost:** Use `1024x1024` for maximum quality
+- Higher resolution = better detail but slower processing
+
+#### **face_enhancer** ⭐ (Recommended)
+- Improves face quality and details after swapping
+- **Best Models:** `gfpgan_1.4`, `restoreformer_plus_plus`
+- **Blend:** 80-100 for strong enhancement
+- **Weight:** Adjust for different model variants
+- Use together with face_swapper for professional results
+
+#### **lip_syncer**
+- Synchronizes lips to audio file
+- **Requirements:** Source audio file must be uploaded
+- **Best Model:** `wav2lip_gan_96` for quality
+- **Weight:** 1.0 for full sync, lower to blend with original
+
+#### **age_modifier**
+- Makes faces younger or older
+- **Direction:** Negative = younger, Positive = older
+- Range: -100 (very young) to +100 (very old)
+
+#### **expression_restorer**
+- Restores target's original facial expressions
+- **Factor:** 100 = full target expression, 0 = source expression
+- Useful to maintain natural emotions after face swap
+
+#### **frame_enhancer**
+- Upscales entire frame (not just face)
+- **Models:** `real_esrgan_x4` (4x upscale), `ultra_sharp_x4` (sharper)
+- Use for low-resolution videos
+- Very slow - use only when needed
+
+#### **frame_colorizer**
+- Colorizes black & white videos/images
+- Multiple artistic styles available
+
+#### **face_editor**
+- Manually adjust facial features
+- Control eyes, mouth, head rotation, expressions
+- Advanced feature for fine-tuning
+
+#### **face_debugger**
+- Shows detection boxes, landmarks, scores
+- Useful for troubleshooting detection issues
+
+---
+
+## Face Detection & Selection
+
+### FACE DETECTOR
+**Purpose:** Detects faces in images/videos for processing.
+
+#### Face Detector Model
+- **yolo_face:** Recommended - best accuracy and speed
+- **retinaface:** Good alternative
+
+#### Face Detector Size
+- **640x640:** Balanced speed and accuracy (recommended)
+- **320x320:** Faster but may miss faces
+- **1280x1280:** Best accuracy but slower
+
+#### Face Detector Angles
+- Enable to detect rotated/tilted faces
+- More angles = better detection but slower
+- Use when faces aren't upright
+
+#### Face Detector Score
+- Confidence threshold (0-1)
+- **0.5:** Standard - good balance
+- Higher = stricter detection, fewer false positives
+- Lower = detect more faces but more false positives
+
+### FACE LANDMARKER
+**Purpose:** Detects facial landmarks (eyes, nose, mouth) for accurate alignment.
+
+#### Face Landmarker Model
+- Detects 5 or 68 facial points
+- Essential for proper face alignment and swapping
+
+#### Face Landmarker Score
+- Confidence threshold (0-1)
+- **0.5:** Generally works well
+- Higher = more accurate landmark detection required
+
+### FACE SELECTOR MODE
+**Purpose:** Choose which faces to process in the target.
+
+#### Modes:
+- **One:** Process first detected face only
+- **Many:** Process all detected faces
+- **Reference:** Track specific face across video frames (best for videos)
+- **Age/Gender/Race filters:** Target specific demographics
+
+#### Reference Face Distance
+- Similarity threshold for reference tracking
+- Lower = stricter matching (same person)
+- Higher = more lenient matching
+
+**Tips:**
+- Use **Reference** mode for videos with multiple people
+- Use **One** for single-person content
+- Use filters to target specific faces in multi-person scenes
+
+---
+
+## Face Masking
+
+### PURPOSE
+Control which parts of the face are swapped and how they blend.
+
+### Face Mask Types
+
+#### **Box**
+- Simple rectangular mask around face
+- **Blur:** Controls edge softness (0.3-0.5 recommended)
+- **Padding:** Expand mask in each direction (top, right, bottom, left)
+- Fast and simple
+
+#### **Occlusion**
+- Avoids occluded areas (glasses, hands, hair)
+- Uses face occluder model
+- More natural when face is partially covered
+
+#### **Region**
+- Masks specific facial regions
+- Uses face parser model
+- Select regions: eyes, nose, mouth, skin, etc.
+
+#### **Area**
+- Masks by facial areas
+- Combine multiple for custom masking
+
+**Tips:**
+- Combine mask types for best results
+- Increase blur for smoother blending
+- Adjust padding if face edges are visible
+
+---
+
+## Output Settings
+
+### IMAGE OUTPUT
+
+#### Output Image Quality (0-100)
+- JPEG compression quality
+- **90-95:** Recommended for high quality
+- **100:** Maximum quality (larger file)
+- **70-80:** Good quality, smaller file
+
+#### Output Image Resolution
+- Can upscale or downscale from original
+- Match source resolution for best quality
+- Upscaling beyond 2x may look artificial
+
+### VIDEO OUTPUT
+
+#### Output Video Encoder
+- **libx264:** Widely compatible, good quality
+- **libx265/hevc:** Better compression, smaller files
+- **h264_nvenc:** GPU-accelerated (NVIDIA only)
+- **copy:** Preserve original encoding
+
+#### Output Video Preset
+- **ultrafast:** Quick but large file
+- **fast/medium:** Balanced
+- **slow/slower:** Best quality and compression (recommended)
+- **veryslow:** Maximum quality, very slow encoding
+
+#### Output Video Quality (0-100)
+- **90-95:** Recommended for professional results
+- **80-85:** Good quality, reasonable file size
+- Higher = better visual quality, larger files
+
+#### Output Video Resolution
+- Can upscale or downscale
+- Higher resolution requires more processing time
+- Match original for best quality/performance ratio
+
+#### Output Video FPS
+- **24:** Cinematic look
+- **30:** Standard video
+- **60:** Smooth motion
+- Match original video FPS for best results
+
+### AUDIO OUTPUT (for videos)
+
+#### Output Audio Encoder
+- **aac:** Widely compatible, good quality (recommended)
+- **libmp3lame:** MP3 format
+- **copy:** Preserve original audio
+
+#### Output Audio Quality (0-100)
+- **80-90:** CD quality
+- **100:** Lossless
+- Higher = better sound, larger file
+
+#### Output Audio Volume (0-200%)
+- **100:** Original volume
+- **<100:** Quieter
+- **>100:** Louder (may cause distortion)
+
+---
+
+## Execution Settings
+
+### EXECUTION PROVIDERS
+**Purpose:** Choose hardware acceleration for processing.
+
+#### Options:
+- **CUDAExecutionProvider:** NVIDIA GPU acceleration (fastest)
+- **CoreMLExecutionProvider:** Apple Silicon acceleration
+- **CPUExecutionProvider:** CPU only (slowest but always available)
+
+**Tips:**
+- Use GPU providers when available for 10-50x speedup
+- CPU is very slow but works on any system
+- Some models require specific providers
+
+### EXECUTION THREAD COUNT
+**Purpose:** Number of parallel processing threads.
+
+**Recommendations:**
+- Set to your CPU core count for optimal performance
+- Higher = faster but uses more CPU/GPU
+- Lower if system becomes unresponsive
+
+### EXECUTION QUEUE COUNT
+**Purpose:** Frames each thread processes before returning.
+
+**Recommendations:**
+- **1-2:** Recommended for most cases
+- Higher = better GPU utilization but more VRAM needed
+- Lower = less memory usage
+
+---
+
+## Memory Management
+
+### VIDEO MEMORY STRATEGY
+**Purpose:** Balance processing speed vs VRAM usage.
+
+#### Options:
+- **Strict:** Low memory usage, slower processing
+- **Moderate:** Balanced (recommended)
+- **Tolerant:** Faster but uses more VRAM
+
+**Tips:**
+- Use Strict if you get out-of-memory errors
+- Use Tolerant if you have high-end GPU (12GB+ VRAM)
+
+### SYSTEM MEMORY LIMIT
+**Purpose:** Limit RAM usage during processing.
+
+- **0:** No limit
+- Set value (in GB) to prevent system crashes
+- Useful for systems with limited RAM
+
+---
+
+## Tips for Best Results
+
+### 🌟 Quality Settings (Best Quality)
+```
+Processors: face_swapper + face_enhancer
+Face Swapper Model: inswapper_128
+Pixel Boost: 1024x1024
+Face Enhancer Model: gfpgan_1.4
+Face Enhancer Blend: 80-100
+Output Image/Video Quality: 90-95
+Video Preset: slow or slower
+```
+
+### ⚡ Speed Settings (Faster Processing)
+```
+Processors: face_swapper only
+Face Swapper Model: inswapper_128
+Pixel Boost: 512x512 or 768x768
+Skip face_enhancer
+Output Quality: 80-85
+Video Preset: medium or fast
+Execution Threads: Max CPU cores
+```
+
+### 🎯 Troubleshooting
+
+#### Face Not Detected
+- Check face detector score (try lowering to 0.3)
+- Enable more detector angles
+- Increase detector size to 1280x1280
+- Ensure face is visible and well-lit
+
+#### Poor Swap Quality
+- Increase pixel boost to 1024x1024
+- Add face_enhancer processor
+- Use higher output quality (90-95)
+- Ensure source and target faces are similar angles
+
+#### Out of Memory Error
+- Lower pixel boost to 512x512 or 768x768
+- Set video memory strategy to "strict"
+- Reduce execution queue count to 1
+- Lower output resolution
+- Process shorter video segments using trim frame
+
+#### Slow Processing
+- Use GPU execution provider (CUDA/CoreML)
+- Reduce pixel boost
+- Skip face_enhancer for faster processing
+- Lower execution thread count
+- Use faster video preset (medium/fast)
+
+#### Unnatural Blending
+- Increase face mask blur (0.4-0.6)
+- Adjust face mask padding
+- Enable occlusion mask type
+- Lower face enhancer blend
+
+---
+
+## Workflow Examples
+
+### Example 1: High-Quality Photo Face Swap
+1. Upload high-resolution source face image
+2. Upload target photo
+3. Select: face_swapper + face_enhancer
+4. Settings:
+   - Face Swapper: inswapper_128, 1024x1024
+   - Face Enhancer: gfpgan_1.4, blend 90
+   - Output Quality: 95
+5. Preview result
+6. Process
+
+### Example 2: Video Face Swap (Multiple People)
+1. Upload source face
+2. Upload target video
+3. Select: face_swapper + face_enhancer
+4. Face Selector: Reference mode
+5. Click reference face in gallery
+6. Settings:
+   - Pixel boost: 1024x1024
+   - Video quality: 90
+   - Preset: slow
+7. Use trim frame to process test segment first
+8. Process full video
+
+### Example 3: Lip Sync Video
+1. Upload source audio (speech/song)
+2. Upload target video
+3. Select: lip_syncer + face_swapper (optional)
+4. Settings:
+   - Lip Syncer: wav2lip_gan_96
+   - Weight: 1.0
+5. Process
+
+---
+
+## Summary Table
+
+| Feature | Recommended Setting | Purpose |
+|---------|-------------------|---------|
+| Face Swapper Model | inswapper_128 | Best quality swapping |
+| Pixel Boost | 1024x1024 | Maximum detail |
+| Face Enhancer | gfpgan_1.4, blend 80 | Improve quality |
+| Output Quality | 90-95 | Near-lossless |
+| Video Preset | slow/slower | Best compression |
+| Execution Provider | CUDA/CoreML | GPU acceleration |
+| Face Selector | Reference (videos) | Track specific person |
+| Face Mask Blur | 0.3-0.5 | Natural blending |
+
+---
+
+**Last Updated:** October 6, 2025
+
+For more information, visit the official FaceFusion documentation.

app.py

@@ -0,0 +1,18 @@
+#!/usr/bin/env python3
+
+import os
+import sys
+
+os.environ['OMP_NUM_THREADS'] = '2'         # Adjust based on CPU cores (e.g., 4 for quad-core)
+os.environ['OPENBLAS_NUM_THREADS'] = '1'    # Prevents thread contention
+os.environ['MKL_NUM_THREADS'] = '1'         # For Intel MKL libraries
+
+from facefusion import core
+
+if __name__ == '__main__':
+    if len(sys.argv) == 1:
+        sys.argv.extend([
+            "run",
+            "--execution-providers", "cpu",         # Force CPU-only mode
+        ])
+    core.cli()

facefusion.ini

@@ -0,0 +1,123 @@
+[paths]
+temp_path =
+jobs_path =
+source_paths =
+target_path =
+output_path =
+
+[patterns]
+source_pattern =
+target_pattern =
+output_pattern =
+
+[face_detector]
+face_detector_model =
+face_detector_size =
+face_detector_angles =
+face_detector_score =
+
+[face_landmarker]
+face_landmarker_model =
+face_landmarker_score =
+
+[face_selector]
+face_selector_mode =
+face_selector_order =
+face_selector_age_start =
+face_selector_age_end =
+face_selector_gender =
+face_selector_race =
+reference_face_position =
+reference_face_distance =
+reference_frame_number =
+
+[face_masker]
+face_occluder_model =
+face_parser_model =
+face_mask_types =
+face_mask_areas =
+face_mask_regions =
+face_mask_blur =
+face_mask_padding =
+
+[frame_extraction]
+trim_frame_start =
+trim_frame_end =
+temp_frame_format =
+keep_temp =
+
+[output_creation]
+output_image_quality = 95
+output_image_resolution =
+output_audio_encoder =
+output_audio_quality =
+output_audio_volume =
+output_video_encoder =
+output_video_preset = slow
+output_video_quality = 95
+output_video_resolution =
+output_video_fps =
+
+[processors]
+processors = face_swapper face_enhancer
+age_modifier_model =
+age_modifier_direction =
+deep_swapper_model =
+deep_swapper_morph =
+expression_restorer_model =
+expression_restorer_factor =
+face_debugger_items =
+face_editor_model =
+face_editor_eyebrow_direction =
+face_editor_eye_gaze_horizontal =
+face_editor_eye_gaze_vertical =
+face_editor_eye_open_ratio =
+face_editor_lip_open_ratio =
+face_editor_mouth_grim =
+face_editor_mouth_pout =
+face_editor_mouth_purse =
+face_editor_mouth_smile =
+face_editor_mouth_position_horizontal =
+face_editor_mouth_position_vertical =
+face_editor_head_pitch =
+face_editor_head_yaw =
+face_editor_head_roll =
+face_enhancer_model = gfpgan_1.4
+face_enhancer_blend = 80
+face_enhancer_weight =
+face_swapper_model = inswapper_128
+face_swapper_pixel_boost = 1024x1024
+frame_colorizer_model =
+frame_colorizer_size =
+frame_colorizer_blend =
+frame_enhancer_model =
+frame_enhancer_blend =
+lip_syncer_model =
+lip_syncer_weight =
+
+[uis]
+open_browser =
+ui_layouts =
+ui_workflow =
+
+[download]
+download_providers =
+download_scope =
+
+[benchmark]
+benchmark_resolutions =
+benchmark_cycle_count =
+
+[execution]
+execution_device_id =
+execution_providers =
+execution_thread_count =
+execution_queue_count =
+
+[memory]
+video_memory_strategy =
+system_memory_limit =
+
+[misc]
+log_level =
+halt_on_error =

facefusion.py

@@ -0,0 +1,10 @@
+#!/usr/bin/env python3
+
+import os
+
+os.environ['OMP_NUM_THREADS'] = '1'
+
+from facefusion import core
+
+if __name__ == '__main__':
+	core.cli()

facefusion/app_context.py

@@ -0,0 +1,16 @@
+import os
+import sys
+
+from facefusion.types import AppContext
+
+
+def detect_app_context() -> AppContext:
+	frame = sys._getframe(1)
+
+	while frame:
+		if os.path.join('facefusion', 'jobs') in frame.f_code.co_filename:
+			return 'cli'
+		if os.path.join('facefusion', 'uis') in frame.f_code.co_filename:
+			return 'ui'
+		frame = frame.f_back
+	return 'cli'

facefusion/args.py

@@ -0,0 +1,140 @@
+from facefusion import state_manager
+from facefusion.filesystem import get_file_name, is_image, is_video, resolve_file_paths
+from facefusion.jobs import job_store
+from facefusion.normalizer import normalize_fps, normalize_padding
+from facefusion.processors.core import get_processors_modules
+from facefusion.types import ApplyStateItem, Args
+from facefusion.vision import create_image_resolutions, create_video_resolutions, detect_image_resolution, detect_video_fps, detect_video_resolution, pack_resolution
+
+
+def reduce_step_args(args : Args) -> Args:
+	step_args =\
+	{
+		key: args[key] for key in args if key in job_store.get_step_keys()
+	}
+	return step_args
+
+
+def reduce_job_args(args : Args) -> Args:
+	job_args =\
+	{
+		key: args[key] for key in args if key in job_store.get_job_keys()
+	}
+	return job_args
+
+
+def collect_step_args() -> Args:
+	step_args =\
+	{
+		key: state_manager.get_item(key) for key in job_store.get_step_keys() #type:ignore[arg-type]
+	}
+	return step_args
+
+
+def collect_job_args() -> Args:
+	job_args =\
+	{
+		key: state_manager.get_item(key) for key in job_store.get_job_keys() #type:ignore[arg-type]
+	}
+	return job_args
+
+
+def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
+	# general
+	apply_state_item('command', args.get('command'))
+	# paths
+	apply_state_item('temp_path', args.get('temp_path'))
+	apply_state_item('jobs_path', args.get('jobs_path'))
+	apply_state_item('source_paths', args.get('source_paths'))
+	apply_state_item('target_path', args.get('target_path'))
+	apply_state_item('output_path', args.get('output_path'))
+	# patterns
+	apply_state_item('source_pattern', args.get('source_pattern'))
+	apply_state_item('target_pattern', args.get('target_pattern'))
+	apply_state_item('output_pattern', args.get('output_pattern'))
+	# face detector
+	apply_state_item('face_detector_model', args.get('face_detector_model'))
+	apply_state_item('face_detector_size', args.get('face_detector_size'))
+	apply_state_item('face_detector_angles', args.get('face_detector_angles'))
+	apply_state_item('face_detector_score', args.get('face_detector_score'))
+	# face landmarker
+	apply_state_item('face_landmarker_model', args.get('face_landmarker_model'))
+	apply_state_item('face_landmarker_score', args.get('face_landmarker_score'))
+	# face selector
+	apply_state_item('face_selector_mode', args.get('face_selector_mode'))
+	apply_state_item('face_selector_order', args.get('face_selector_order'))
+	apply_state_item('face_selector_age_start', args.get('face_selector_age_start'))
+	apply_state_item('face_selector_age_end', args.get('face_selector_age_end'))
+	apply_state_item('face_selector_gender', args.get('face_selector_gender'))
+	apply_state_item('face_selector_race', args.get('face_selector_race'))
+	apply_state_item('reference_face_position', args.get('reference_face_position'))
+	apply_state_item('reference_face_distance', args.get('reference_face_distance'))
+	apply_state_item('reference_frame_number', args.get('reference_frame_number'))
+	# face masker
+	apply_state_item('face_occluder_model', args.get('face_occluder_model'))
+	apply_state_item('face_parser_model', args.get('face_parser_model'))
+	apply_state_item('face_mask_types', args.get('face_mask_types'))
+	apply_state_item('face_mask_areas', args.get('face_mask_areas'))
+	apply_state_item('face_mask_regions', args.get('face_mask_regions'))
+	apply_state_item('face_mask_blur', args.get('face_mask_blur'))
+	apply_state_item('face_mask_padding', normalize_padding(args.get('face_mask_padding')))
+	# frame extraction
+	apply_state_item('trim_frame_start', args.get('trim_frame_start'))
+	apply_state_item('trim_frame_end', args.get('trim_frame_end'))
+	apply_state_item('temp_frame_format', args.get('temp_frame_format'))
+	apply_state_item('keep_temp', args.get('keep_temp'))
+	# output creation
+	apply_state_item('output_image_quality', args.get('output_image_quality'))
+	if is_image(args.get('target_path')):
+		output_image_resolution = detect_image_resolution(args.get('target_path'))
+		output_image_resolutions = create_image_resolutions(output_image_resolution)
+		if args.get('output_image_resolution') in output_image_resolutions:
+			apply_state_item('output_image_resolution', args.get('output_image_resolution'))
+		else:
+			apply_state_item('output_image_resolution', pack_resolution(output_image_resolution))
+	apply_state_item('output_audio_encoder', args.get('output_audio_encoder'))
+	apply_state_item('output_audio_quality', args.get('output_audio_quality'))
+	apply_state_item('output_audio_volume', args.get('output_audio_volume'))
+	apply_state_item('output_video_encoder', args.get('output_video_encoder'))
+	apply_state_item('output_video_preset', args.get('output_video_preset'))
+	apply_state_item('output_video_quality', args.get('output_video_quality'))
+	if is_video(args.get('target_path')):
+		output_video_resolution = detect_video_resolution(args.get('target_path'))
+		output_video_resolutions = create_video_resolutions(output_video_resolution)
+		if args.get('output_video_resolution') in output_video_resolutions:
+			apply_state_item('output_video_resolution', args.get('output_video_resolution'))
+		else:
+			apply_state_item('output_video_resolution', pack_resolution(output_video_resolution))
+	if args.get('output_video_fps') or is_video(args.get('target_path')):
+		output_video_fps = normalize_fps(args.get('output_video_fps')) or detect_video_fps(args.get('target_path'))
+		apply_state_item('output_video_fps', output_video_fps)
+	# processors
+	available_processors = [ get_file_name(file_path) for file_path in resolve_file_paths('facefusion/processors/modules') ]
+	apply_state_item('processors', args.get('processors'))
+	for processor_module in get_processors_modules(available_processors):
+		processor_module.apply_args(args, apply_state_item)
+	# uis
+	apply_state_item('open_browser', args.get('open_browser'))
+	apply_state_item('ui_layouts', args.get('ui_layouts'))
+	apply_state_item('ui_workflow', args.get('ui_workflow'))
+	# execution
+	apply_state_item('execution_device_id', args.get('execution_device_id'))
+	apply_state_item('execution_providers', args.get('execution_providers'))
+	apply_state_item('execution_thread_count', args.get('execution_thread_count'))
+	apply_state_item('execution_queue_count', args.get('execution_queue_count'))
+	# download
+	apply_state_item('download_providers', args.get('download_providers'))
+	apply_state_item('download_scope', args.get('download_scope'))
+	# benchmark
+	apply_state_item('benchmark_resolutions', args.get('benchmark_resolutions'))
+	apply_state_item('benchmark_cycle_count', args.get('benchmark_cycle_count'))
+	# memory
+	apply_state_item('video_memory_strategy', args.get('video_memory_strategy'))
+	apply_state_item('system_memory_limit', args.get('system_memory_limit'))
+	# misc
+	apply_state_item('log_level', args.get('log_level'))
+	apply_state_item('halt_on_error', args.get('halt_on_error'))
+	# jobs
+	apply_state_item('job_id', args.get('job_id'))
+	apply_state_item('job_status', args.get('job_status'))
+	apply_state_item('step_index', args.get('step_index'))

facefusion/audio.py

@@ -0,0 +1,143 @@
+from functools import lru_cache
+from typing import Any, List, Optional
+
+import numpy
+import scipy
+from numpy.typing import NDArray
+
+from facefusion.ffmpeg import read_audio_buffer
+from facefusion.filesystem import is_audio
+from facefusion.types import Audio, AudioFrame, Fps, Mel, MelFilterBank, Spectrogram
+from facefusion.voice_extractor import batch_extract_voice
+
+
+@lru_cache()
+def read_static_audio(audio_path : str, fps : Fps) -> Optional[List[AudioFrame]]:
+	return read_audio(audio_path, fps)
+
+
+def read_audio(audio_path : str, fps : Fps) -> Optional[List[AudioFrame]]:
+	audio_sample_rate = 48000
+	audio_sample_size = 16
+	audio_channel_total = 2
+
+	if is_audio(audio_path):
+		audio_buffer = read_audio_buffer(audio_path, audio_sample_rate, audio_sample_size, audio_channel_total)
+		audio = numpy.frombuffer(audio_buffer, dtype = numpy.int16).reshape(-1, 2)
+		audio = prepare_audio(audio)
+		spectrogram = create_spectrogram(audio)
+		audio_frames = extract_audio_frames(spectrogram, fps)
+		return audio_frames
+	return None
+
+
+@lru_cache()
+def read_static_voice(audio_path : str, fps : Fps) -> Optional[List[AudioFrame]]:
+	return read_voice(audio_path, fps)
+
+
+def read_voice(audio_path : str, fps : Fps) -> Optional[List[AudioFrame]]:
+	voice_sample_rate = 48000
+	voice_sample_size = 16
+	voice_channel_total = 2
+	voice_chunk_size = 240 * 1024
+	voice_step_size = 180 * 1024
+
+	if is_audio(audio_path):
+		audio_buffer = read_audio_buffer(audio_path, voice_sample_rate, voice_sample_size, voice_channel_total)
+		audio = numpy.frombuffer(audio_buffer, dtype = numpy.int16).reshape(-1, 2)
+		audio = batch_extract_voice(audio, voice_chunk_size, voice_step_size)
+		audio = prepare_voice(audio)
+		spectrogram = create_spectrogram(audio)
+		audio_frames = extract_audio_frames(spectrogram, fps)
+		return audio_frames
+	return None
+
+
+def get_audio_frame(audio_path : str, fps : Fps, frame_number : int = 0) -> Optional[AudioFrame]:
+	if is_audio(audio_path):
+		audio_frames = read_static_audio(audio_path, fps)
+		if frame_number in range(len(audio_frames)):
+			return audio_frames[frame_number]
+	return None
+
+
+def extract_audio_frames(spectrogram : Spectrogram, fps : Fps) -> List[AudioFrame]:
+	audio_frames = []
+	mel_filter_total = 80
+	audio_step_size = 16
+	indices = numpy.arange(0, spectrogram.shape[1], mel_filter_total / fps).astype(numpy.int16)
+	indices = indices[indices >= audio_step_size]
+
+	for index in indices:
+		start = max(0, index - audio_step_size)
+		audio_frames.append(spectrogram[:, start:index])
+
+	return audio_frames
+
+
+def get_voice_frame(audio_path : str, fps : Fps, frame_number : int = 0) -> Optional[AudioFrame]:
+	if is_audio(audio_path):
+		voice_frames = read_static_voice(audio_path, fps)
+		if frame_number in range(len(voice_frames)):
+			return voice_frames[frame_number]
+	return None
+
+
+def create_empty_audio_frame() -> AudioFrame:
+	mel_filter_total = 80
+	audio_step_size = 16
+	audio_frame = numpy.zeros((mel_filter_total, audio_step_size)).astype(numpy.int16)
+	return audio_frame
+
+
+def prepare_audio(audio : Audio) -> Audio:
+	if audio.ndim > 1:
+		audio = numpy.mean(audio, axis = 1)
+	audio = audio / numpy.max(numpy.abs(audio), axis = 0)
+	audio = scipy.signal.lfilter([ 1.0, -0.97 ], [ 1.0 ], audio)
+	return audio
+
+
+def prepare_voice(audio : Audio) -> Audio:
+	audio_sample_rate = 48000
+	audio_resample_rate = 16000
+	audio_resample_factor = round(len(audio) * audio_resample_rate / audio_sample_rate)
+	audio = scipy.signal.resample(audio, audio_resample_factor)
+	audio = prepare_audio(audio)
+	return audio
+
+
+def convert_hertz_to_mel(hertz : float) -> float:
+	return 2595 * numpy.log10(1 + hertz / 700)
+
+
+def convert_mel_to_hertz(mel : Mel) -> NDArray[Any]:
+	return 700 * (10 ** (mel / 2595) - 1)
+
+
+def create_mel_filter_bank() -> MelFilterBank:
+	audio_sample_rate = 16000
+	audio_min_frequency = 55.0
+	audio_max_frequency = 7600.0
+	mel_filter_total = 80
+	mel_bin_total = 800
+	mel_filter_bank = numpy.zeros((mel_filter_total, mel_bin_total // 2 + 1))
+	mel_frequency_range = numpy.linspace(convert_hertz_to_mel(audio_min_frequency), convert_hertz_to_mel(audio_max_frequency), mel_filter_total + 2)
+	indices = numpy.floor((mel_bin_total + 1) * convert_mel_to_hertz(mel_frequency_range) / audio_sample_rate).astype(numpy.int16)
+
+	for index in range(mel_filter_total):
+		start = indices[index]
+		end = indices[index + 1]
+		mel_filter_bank[index, start:end] = scipy.signal.windows.triang(end - start)
+
+	return mel_filter_bank
+
+
+def create_spectrogram(audio : Audio) -> Spectrogram:
+	mel_bin_total = 800
+	mel_bin_overlap = 600
+	mel_filter_bank = create_mel_filter_bank()
+	spectrogram = scipy.signal.stft(audio, nperseg = mel_bin_total, nfft = mel_bin_total, noverlap = mel_bin_overlap)[2]
+	spectrogram = numpy.dot(mel_filter_bank, numpy.abs(spectrogram))
+	return spectrogram

facefusion/benchmarker.py

@@ -0,0 +1,106 @@
+import hashlib
+import os
+import statistics
+import tempfile
+from time import perf_counter
+from typing import Generator, List
+
+import facefusion.choices
+from facefusion import core, state_manager
+from facefusion.cli_helper import render_table
+from facefusion.download import conditional_download, resolve_download_url
+from facefusion.filesystem import get_file_extension
+from facefusion.types import BenchmarkCycleSet
+from facefusion.vision import count_video_frame_total, detect_video_fps, detect_video_resolution, pack_resolution
+
+
+def pre_check() -> bool:
+	conditional_download('.assets/examples',
+	[
+		resolve_download_url('examples-3.0.0', 'source.jpg'),
+		resolve_download_url('examples-3.0.0', 'source.mp3'),
+		resolve_download_url('examples-3.0.0', 'target-240p.mp4'),
+		resolve_download_url('examples-3.0.0', 'target-360p.mp4'),
+		resolve_download_url('examples-3.0.0', 'target-540p.mp4'),
+		resolve_download_url('examples-3.0.0', 'target-720p.mp4'),
+		resolve_download_url('examples-3.0.0', 'target-1080p.mp4'),
+		resolve_download_url('examples-3.0.0', 'target-1440p.mp4'),
+		resolve_download_url('examples-3.0.0', 'target-2160p.mp4')
+	])
+	return True
+
+
+def run() -> Generator[List[BenchmarkCycleSet], None, None]:
+	benchmark_resolutions = state_manager.get_item('benchmark_resolutions')
+	benchmark_cycle_count = state_manager.get_item('benchmark_cycle_count')
+
+	state_manager.init_item('source_paths', [ '.assets/examples/source.jpg', '.assets/examples/source.mp3' ])
+	state_manager.init_item('face_landmarker_score', 0)
+	state_manager.init_item('temp_frame_format', 'bmp')
+	state_manager.init_item('output_audio_volume', 0)
+	state_manager.init_item('output_video_preset', 'ultrafast')
+	state_manager.init_item('video_memory_strategy', 'tolerant')
+
+	benchmarks = []
+	target_paths = [facefusion.choices.benchmark_set.get(benchmark_resolution) for benchmark_resolution in benchmark_resolutions if benchmark_resolution in facefusion.choices.benchmark_set]
+
+	for target_path in target_paths:
+		state_manager.set_item('target_path', target_path)
+		state_manager.set_item('output_path', suggest_output_path(state_manager.get_item('target_path')))
+		benchmarks.append(cycle(benchmark_cycle_count))
+		yield benchmarks
+
+
+def cycle(cycle_count : int) -> BenchmarkCycleSet:
+	process_times = []
+	video_frame_total = count_video_frame_total(state_manager.get_item('target_path'))
+	output_video_resolution = detect_video_resolution(state_manager.get_item('target_path'))
+	state_manager.set_item('output_video_resolution', pack_resolution(output_video_resolution))
+	state_manager.set_item('output_video_fps', detect_video_fps(state_manager.get_item('target_path')))
+
+	core.conditional_process()
+
+	for index in range(cycle_count):
+		start_time = perf_counter()
+		core.conditional_process()
+		end_time = perf_counter()
+		process_times.append(end_time - start_time)
+
+	average_run = round(statistics.mean(process_times), 2)
+	fastest_run = round(min(process_times), 2)
+	slowest_run = round(max(process_times), 2)
+	relative_fps = round(video_frame_total * cycle_count / sum(process_times), 2)
+
+	return\
+	{
+		'target_path': state_manager.get_item('target_path'),
+		'cycle_count': cycle_count,
+		'average_run': average_run,
+		'fastest_run': fastest_run,
+		'slowest_run': slowest_run,
+		'relative_fps': relative_fps
+	}
+
+
+def suggest_output_path(target_path : str) -> str:
+	target_file_extension = get_file_extension(target_path)
+	return os.path.join(tempfile.gettempdir(), hashlib.sha1().hexdigest()[:8] + target_file_extension)
+
+
+def render() -> None:
+	benchmarks = []
+	headers =\
+	[
+		'target_path',
+		'cycle_count',
+		'average_run',
+		'fastest_run',
+		'slowest_run',
+		'relative_fps'
+	]
+
+	for benchmark in run():
+		benchmarks = benchmark
+
+	contents = [ list(benchmark_set.values()) for benchmark_set in benchmarks ]
+	render_table(headers, contents)

facefusion/choices.py

@@ -0,0 +1,165 @@
+import logging
+from typing import List, Sequence
+
+from facefusion.common_helper import create_float_range, create_int_range
+from facefusion.types import Angle, AudioEncoder, AudioFormat, AudioTypeSet, BenchmarkResolution, BenchmarkSet, DownloadProvider, DownloadProviderSet, DownloadScope, EncoderSet, ExecutionProvider, ExecutionProviderSet, FaceDetectorModel, FaceDetectorSet, FaceLandmarkerModel, FaceMaskArea, FaceMaskAreaSet, FaceMaskRegion, FaceMaskRegionSet, FaceMaskType, FaceOccluderModel, FaceParserModel, FaceSelectorMode, FaceSelectorOrder, Gender, ImageFormat, ImageTypeSet, JobStatus, LogLevel, LogLevelSet, Race, Score, TempFrameFormat, UiWorkflow, VideoEncoder, VideoFormat, VideoMemoryStrategy, VideoPreset, VideoTypeSet, WebcamMode
+
+face_detector_set : FaceDetectorSet =\
+{
+	'many': [ '640x640' ],
+	'retinaface': [ '160x160', '320x320', '480x480', '512x512', '640x640' ],
+	'scrfd': [ '160x160', '320x320', '480x480', '512x512', '640x640' ],
+	'yolo_face': [ '640x640' ]
+}
+face_detector_models : List[FaceDetectorModel] = list(face_detector_set.keys())
+face_landmarker_models : List[FaceLandmarkerModel] = [ 'many', '2dfan4', 'peppa_wutz' ]
+face_selector_modes : List[FaceSelectorMode] = [ 'many', 'one', 'reference' ]
+face_selector_orders : List[FaceSelectorOrder] = [ 'left-right', 'right-left', 'top-bottom', 'bottom-top', 'small-large', 'large-small', 'best-worst', 'worst-best' ]
+face_selector_genders : List[Gender] = [ 'female', 'male' ]
+face_selector_races : List[Race] = [ 'white', 'black', 'latino', 'asian', 'indian', 'arabic' ]
+face_occluder_models : List[FaceOccluderModel] = [ 'xseg_1', 'xseg_2', 'xseg_3' ]
+face_parser_models : List[FaceParserModel] = [ 'bisenet_resnet_18', 'bisenet_resnet_34' ]
+face_mask_types : List[FaceMaskType] = [ 'box', 'occlusion', 'area', 'region' ]
+face_mask_area_set : FaceMaskAreaSet =\
+{
+	'upper-face': [ 0, 1, 2, 31, 32, 33, 34, 35, 14, 15, 16, 26, 25, 24, 23, 22, 21, 20, 19, 18, 17 ],
+	'lower-face': [ 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 35, 34, 33, 32, 31 ],
+	'mouth': [ 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67 ]
+}
+face_mask_region_set : FaceMaskRegionSet =\
+{
+	'skin': 1,
+	'left-eyebrow': 2,
+	'right-eyebrow': 3,
+	'left-eye': 4,
+	'right-eye': 5,
+	'glasses': 6,
+	'nose': 10,
+	'mouth': 11,
+	'upper-lip': 12,
+	'lower-lip': 13
+}
+face_mask_areas : List[FaceMaskArea] = list(face_mask_area_set.keys())
+face_mask_regions : List[FaceMaskRegion] = list(face_mask_region_set.keys())
+
+audio_type_set : AudioTypeSet =\
+{
+	'flac': 'audio/flac',
+	'm4a': 'audio/mp4',
+	'mp3': 'audio/mpeg',
+	'ogg': 'audio/ogg',
+	'opus': 'audio/opus',
+	'wav': 'audio/x-wav'
+}
+image_type_set : ImageTypeSet =\
+{
+	'bmp': 'image/bmp',
+	'jpeg': 'image/jpeg',
+	'png': 'image/png',
+	'tiff': 'image/tiff',
+	'webp': 'image/webp'
+}
+video_type_set : VideoTypeSet =\
+{
+	'avi': 'video/x-msvideo',
+	'm4v': 'video/mp4',
+	'mkv': 'video/x-matroska',
+	'mp4': 'video/mp4',
+	'mov': 'video/quicktime',
+	'webm': 'video/webm'
+}
+audio_formats : List[AudioFormat] = list(audio_type_set.keys())
+image_formats : List[ImageFormat] = list(image_type_set.keys())
+video_formats : List[VideoFormat] = list(video_type_set.keys())
+temp_frame_formats : List[TempFrameFormat] = [ 'bmp', 'jpeg', 'png', 'tiff' ]
+
+output_encoder_set : EncoderSet =\
+{
+	'audio': [ 'flac', 'aac', 'libmp3lame', 'libopus', 'libvorbis', 'pcm_s16le', 'pcm_s32le' ],
+	'video': [ 'libx264', 'libx265', 'libvpx-vp9', 'h264_nvenc', 'hevc_nvenc', 'h264_amf', 'hevc_amf', 'h264_qsv', 'hevc_qsv', 'h264_videotoolbox', 'hevc_videotoolbox', 'rawvideo' ]
+}
+output_audio_encoders : List[AudioEncoder] = output_encoder_set.get('audio')
+output_video_encoders : List[VideoEncoder] = output_encoder_set.get('video')
+output_video_presets : List[VideoPreset] = [ 'ultrafast', 'superfast', 'veryfast', 'faster', 'fast', 'medium', 'slow', 'slower', 'veryslow' ]
+
+image_template_sizes : List[float] = [ 0.25, 0.5, 0.75, 1, 1.5, 2, 2.5, 3, 3.5, 4 ]
+video_template_sizes : List[int] = [ 240, 360, 480, 540, 720, 1080, 1440, 2160, 4320 ]
+
+benchmark_set : BenchmarkSet =\
+{
+	'240p': '.assets/examples/target-240p.mp4',
+	'360p': '.assets/examples/target-360p.mp4',
+	'540p': '.assets/examples/target-540p.mp4',
+	'720p': '.assets/examples/target-720p.mp4',
+	'1080p': '.assets/examples/target-1080p.mp4',
+	'1440p': '.assets/examples/target-1440p.mp4',
+	'2160p': '.assets/examples/target-2160p.mp4'
+}
+benchmark_resolutions : List[BenchmarkResolution] = list(benchmark_set.keys())
+
+webcam_modes : List[WebcamMode] = [ 'inline', 'udp', 'v4l2' ]
+webcam_resolutions : List[str] = [ '320x240', '640x480', '800x600', '1024x768', '1280x720', '1280x960', '1920x1080', '2560x1440', '3840x2160' ]
+
+execution_provider_set : ExecutionProviderSet =\
+{
+	'cuda': 'CUDAExecutionProvider',
+	'tensorrt': 'TensorrtExecutionProvider',
+	'directml': 'DmlExecutionProvider',
+	'rocm': 'ROCMExecutionProvider',
+	'openvino': 'OpenVINOExecutionProvider',
+	'coreml': 'CoreMLExecutionProvider',
+	'cpu': 'CPUExecutionProvider'
+}
+execution_providers : List[ExecutionProvider] = list(execution_provider_set.keys())
+download_provider_set : DownloadProviderSet =\
+{
+	'github':
+	{
+		'urls':
+		[
+			'https://github.com'
+		],
+		'path': '/facefusion/facefusion-assets/releases/download/{base_name}/{file_name}'
+	},
+	'huggingface':
+	{
+		'urls':
+		[
+			'https://huggingface.co',
+			'https://hf-mirror.com'
+		],
+		'path': '/facefusion/{base_name}/resolve/main/{file_name}'
+	}
+}
+download_providers : List[DownloadProvider] = list(download_provider_set.keys())
+download_scopes : List[DownloadScope] = [ 'lite', 'full' ]
+
+video_memory_strategies : List[VideoMemoryStrategy] = [ 'strict', 'moderate', 'tolerant' ]
+
+log_level_set : LogLevelSet =\
+{
+	'error': logging.ERROR,
+	'warn': logging.WARNING,
+	'info': logging.INFO,
+	'debug': logging.DEBUG
+}
+log_levels : List[LogLevel] = list(log_level_set.keys())
+
+ui_workflows : List[UiWorkflow] = [ 'instant_runner', 'job_runner', 'job_manager' ]
+job_statuses : List[JobStatus] = [ 'drafted', 'queued', 'completed', 'failed' ]
+
+benchmark_cycle_count_range : Sequence[int] = create_int_range(1, 10, 1)
+execution_thread_count_range : Sequence[int] = create_int_range(1, 32, 1)
+execution_queue_count_range : Sequence[int] = create_int_range(1, 4, 1)
+system_memory_limit_range : Sequence[int] = create_int_range(0, 128, 4)
+face_detector_angles : Sequence[Angle] = create_int_range(0, 270, 90)
+face_detector_score_range : Sequence[Score] = create_float_range(0.0, 1.0, 0.05)
+face_landmarker_score_range : Sequence[Score] = create_float_range(0.0, 1.0, 0.05)
+face_mask_blur_range : Sequence[float] = create_float_range(0.0, 1.0, 0.05)
+face_mask_padding_range : Sequence[int] = create_int_range(0, 100, 1)
+face_selector_age_range : Sequence[int] = create_int_range(0, 100, 1)
+reference_face_distance_range : Sequence[float] = create_float_range(0.0, 1.0, 0.05)
+output_image_quality_range : Sequence[int] = create_int_range(0, 100, 1)
+output_audio_quality_range : Sequence[int] = create_int_range(0, 100, 1)
+output_audio_volume_range : Sequence[int] = create_int_range(0, 100, 1)
+output_video_quality_range : Sequence[int] = create_int_range(0, 100, 1)

facefusion/cli_helper.py

@@ -0,0 +1,35 @@
+from typing import Tuple
+
+from facefusion.logger import get_package_logger
+from facefusion.types import TableContents, TableHeaders
+
+
+def render_table(headers : TableHeaders, contents : TableContents) -> None:
+	package_logger = get_package_logger()
+	table_column, table_separator = create_table_parts(headers, contents)
+
+	package_logger.critical(table_separator)
+	package_logger.critical(table_column.format(*headers))
+	package_logger.critical(table_separator)
+
+	for content in contents:
+		content = [ str(value) for value in content ]
+		package_logger.critical(table_column.format(*content))
+
+	package_logger.critical(table_separator)
+
+
+def create_table_parts(headers : TableHeaders, contents : TableContents) -> Tuple[str, str]:
+	column_parts = []
+	separator_parts = []
+	widths = [ len(header) for header in headers ]
+
+	for content in contents:
+		for index, value in enumerate(content):
+			widths[index] = max(widths[index], len(str(value)))
+
+	for width in widths:
+		column_parts.append('{:<' + str(width) + '}')
+		separator_parts.append('-' * width)
+
+	return '| ' + ' | '.join(column_parts) + ' |', '+-' + '-+-'.join(separator_parts) + '-+'

facefusion/common_helper.py

@@ -0,0 +1,84 @@
+import platform
+from typing import Any, Iterable, Optional, Reversible, Sequence
+
+
+def is_linux() -> bool:
+	return platform.system().lower() == 'linux'
+
+
+def is_macos() -> bool:
+	return platform.system().lower() == 'darwin'
+
+
+def is_windows() -> bool:
+	return platform.system().lower() == 'windows'
+
+
+def create_int_metavar(int_range : Sequence[int]) -> str:
+	return '[' + str(int_range[0]) + '..' + str(int_range[-1]) + ':' + str(calc_int_step(int_range)) + ']'
+
+
+def create_float_metavar(float_range : Sequence[float]) -> str:
+	return '[' + str(float_range[0]) + '..' + str(float_range[-1]) + ':' + str(calc_float_step(float_range)) + ']'
+
+
+def create_int_range(start : int, end : int, step : int) -> Sequence[int]:
+	int_range = []
+	current = start
+
+	while current <= end:
+		int_range.append(current)
+		current += step
+	return int_range
+
+
+def create_float_range(start : float, end : float, step : float) -> Sequence[float]:
+	float_range = []
+	current = start
+
+	while current <= end:
+		float_range.append(round(current, 2))
+		current = round(current + step, 2)
+	return float_range
+
+
+def calc_int_step(int_range : Sequence[int]) -> int:
+	return int_range[1] - int_range[0]
+
+
+def calc_float_step(float_range : Sequence[float]) -> float:
+	return round(float_range[1] - float_range[0], 2)
+
+
+def cast_int(value : Any) -> Optional[int]:
+	try:
+		return int(value)
+	except (ValueError, TypeError):
+		return None
+
+
+def cast_float(value : Any) -> Optional[float]:
+	try:
+		return float(value)
+	except (ValueError, TypeError):
+		return None
+
+
+def cast_bool(value : Any) -> Optional[bool]:
+	if value == 'True':
+		return True
+	if value == 'False':
+		return False
+	return None
+
+
+def get_first(__list__ : Any) -> Any:
+	if isinstance(__list__, Iterable):
+		return next(iter(__list__), None)
+	return None
+
+
+def get_last(__list__ : Any) -> Any:
+	if isinstance(__list__, Reversible):
+		return next(reversed(__list__), None)
+	return None

facefusion/config.py

@@ -0,0 +1,74 @@
+from configparser import ConfigParser
+from typing import List, Optional
+
+from facefusion import state_manager
+from facefusion.common_helper import cast_bool, cast_float, cast_int
+
+CONFIG_PARSER = None
+
+
+def get_config_parser() -> ConfigParser:
+	global CONFIG_PARSER
+
+	if CONFIG_PARSER is None:
+		CONFIG_PARSER = ConfigParser()
+		CONFIG_PARSER.read(state_manager.get_item('config_path'), encoding = 'utf-8')
+	return CONFIG_PARSER
+
+
+def clear_config_parser() -> None:
+	global CONFIG_PARSER
+
+	CONFIG_PARSER = None
+
+
+def get_str_value(section : str, option : str, fallback : Optional[str] = None) -> Optional[str]:
+	config_parser = get_config_parser()
+
+	if config_parser.has_option(section, option) and config_parser.get(section, option).strip():
+		return config_parser.get(section, option)
+	return fallback
+
+
+def get_int_value(section : str, option : str, fallback : Optional[str] = None) -> Optional[int]:
+	config_parser = get_config_parser()
+
+	if config_parser.has_option(section, option) and config_parser.get(section, option).strip():
+		return config_parser.getint(section, option)
+	return cast_int(fallback)
+
+
+def get_float_value(section : str, option : str, fallback : Optional[str] = None) -> Optional[float]:
+	config_parser = get_config_parser()
+
+	if config_parser.has_option(section, option) and config_parser.get(section, option).strip():
+		return config_parser.getfloat(section, option)
+	return cast_float(fallback)
+
+
+def get_bool_value(section : str, option : str, fallback : Optional[str] = None) -> Optional[bool]:
+	config_parser = get_config_parser()
+
+	if config_parser.has_option(section, option) and config_parser.get(section, option).strip():
+		return config_parser.getboolean(section, option)
+	return cast_bool(fallback)
+
+
+def get_str_list(section : str, option : str, fallback : Optional[str] = None) -> Optional[List[str]]:
+	config_parser = get_config_parser()
+
+	if config_parser.has_option(section, option) and config_parser.get(section, option).strip():
+		return config_parser.get(section, option).split()
+	if fallback:
+		return fallback.split()
+	return None
+
+
+def get_int_list(section : str, option : str, fallback : Optional[str] = None) -> Optional[List[int]]:
+	config_parser = get_config_parser()
+
+	if config_parser.has_option(section, option) and config_parser.get(section, option).strip():
+		return list(map(int, config_parser.get(section, option).split()))
+	if fallback:
+		return list(map(int, fallback.split()))
+	return None

facefusion/content_analyser.py

@@ -0,0 +1,225 @@
+from functools import lru_cache
+from typing import List, Tuple
+
+import numpy
+from tqdm import tqdm
+
+from facefusion import inference_manager, state_manager, wording
+from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
+from facefusion.execution import has_execution_provider
+from facefusion.filesystem import resolve_relative_path
+from facefusion.thread_helper import conditional_thread_semaphore
+from facefusion.types import Detection, DownloadScope, DownloadSet, ExecutionProvider, Fps, InferencePool, ModelSet, VisionFrame
+from facefusion.vision import detect_video_fps, fit_frame, read_image, read_video_frame
+
+STREAM_COUNTER = 0
+
+
+@lru_cache(maxsize = None)
+def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
+	return\
+	{
+		'nsfw_1':
+		{
+			'hashes':
+			{
+				'content_analyser':
+				{
+					'url': resolve_download_url('models-3.3.0', 'nsfw_1.hash'),
+					'path': resolve_relative_path('../.assets/models/nsfw_1.hash')
+				}
+			},
+			'sources':
+			{
+				'content_analyser':
+				{
+					'url': resolve_download_url('models-3.3.0', 'nsfw_1.onnx'),
+					'path': resolve_relative_path('../.assets/models/nsfw_1.onnx')
+				}
+			},
+			'size': (640, 640),
+			'mean': (0.0, 0.0, 0.0),
+			'standard_deviation': (1.0, 1.0, 1.0)
+		},
+		'nsfw_2':
+		{
+			'hashes':
+			{
+				'content_analyser':
+				{
+					'url': resolve_download_url('models-3.3.0', 'nsfw_2.hash'),
+					'path': resolve_relative_path('../.assets/models/nsfw_2.hash')
+				}
+			},
+			'sources':
+			{
+				'content_analyser':
+				{
+					'url': resolve_download_url('models-3.3.0', 'nsfw_2.onnx'),
+					'path': resolve_relative_path('../.assets/models/nsfw_2.onnx')
+				}
+			},
+			'size': (384, 384),
+			'mean': (0.5, 0.5, 0.5),
+			'standard_deviation': (0.5, 0.5, 0.5)
+		},
+		'nsfw_3':
+		{
+			'hashes':
+			{
+				'content_analyser':
+				{
+					'url': resolve_download_url('models-3.3.0', 'nsfw_3.hash'),
+					'path': resolve_relative_path('../.assets/models/nsfw_3.hash')
+				}
+			},
+			'sources':
+			{
+				'content_analyser':
+				{
+					'url': resolve_download_url('models-3.3.0', 'nsfw_3.onnx'),
+					'path': resolve_relative_path('../.assets/models/nsfw_3.onnx')
+				}
+			},
+			'size': (448, 448),
+			'mean': (0.48145466, 0.4578275, 0.40821073),
+			'standard_deviation': (0.26862954, 0.26130258, 0.27577711)
+		}
+	}
+
+
+def get_inference_pool() -> InferencePool:
+	model_names = [ 'nsfw_1', 'nsfw_2', 'nsfw_3' ]
+	_, model_source_set = collect_model_downloads()
+
+	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)
+
+
+def clear_inference_pool() -> None:
+	model_names = [ 'nsfw_1', 'nsfw_2', 'nsfw_3' ]
+	inference_manager.clear_inference_pool(__name__, model_names)
+
+
+def resolve_execution_providers() -> List[ExecutionProvider]:
+	if has_execution_provider('coreml'):
+		return [ 'cpu' ]
+	return state_manager.get_item('execution_providers')
+
+
+def collect_model_downloads() -> Tuple[DownloadSet, DownloadSet]:
+	model_set = create_static_model_set('full')
+	model_hash_set = {}
+	model_source_set = {}
+
+	for content_analyser_model in [ 'nsfw_1', 'nsfw_2', 'nsfw_3' ]:
+		model_hash_set[content_analyser_model] = model_set.get(content_analyser_model).get('hashes').get('content_analyser')
+		model_source_set[content_analyser_model] = model_set.get(content_analyser_model).get('sources').get('content_analyser')
+
+	return model_hash_set, model_source_set
+
+
+def pre_check() -> bool:
+	model_hash_set, model_source_set = collect_model_downloads()
+
+	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)
+
+
+def analyse_stream(vision_frame : VisionFrame, video_fps : Fps) -> bool:
+	global STREAM_COUNTER
+
+	STREAM_COUNTER = STREAM_COUNTER + 1
+	if STREAM_COUNTER % int(video_fps) == 0:
+		return analyse_frame(vision_frame)
+	return False
+
+
+def analyse_frame(vision_frame : VisionFrame) -> bool:
+	return detect_nsfw(vision_frame)
+
+
+@lru_cache(maxsize = None)
+def analyse_image(image_path : str) -> bool:
+	vision_frame = read_image(image_path)
+	return analyse_frame(vision_frame)
+
+
+@lru_cache(maxsize = None)
+def analyse_video(video_path : str, trim_frame_start : int, trim_frame_end : int) -> bool:
+	video_fps = detect_video_fps(video_path)
+	frame_range = range(trim_frame_start, trim_frame_end)
+	rate = 0.0
+	total = 0
+	counter = 0
+
+	with tqdm(total = len(frame_range), desc = wording.get('analysing'), unit = 'frame', ascii = ' =', disable = state_manager.get_item('log_level') in [ 'warn', 'error' ]) as progress:
+
+		for frame_number in frame_range:
+			if frame_number % int(video_fps) == 0:
+				vision_frame = read_video_frame(video_path, frame_number)
+				total += 1
+				if analyse_frame(vision_frame):
+					counter += 1
+			if counter > 0 and total > 0:
+				rate = counter / total * 100
+			progress.set_postfix(rate = rate)
+			progress.update()
+
+	return bool(rate > 10.0)
+
+
+def detect_nsfw(vision_frame : VisionFrame) -> bool:
+	is_nsfw_1 = detect_with_nsfw_1(vision_frame)
+	is_nsfw_2 = detect_with_nsfw_2(vision_frame)
+	is_nsfw_3 = detect_with_nsfw_3(vision_frame)
+
+	return False
+
+
+def detect_with_nsfw_1(vision_frame : VisionFrame) -> bool:
+	detect_vision_frame = prepare_detect_frame(vision_frame, 'nsfw_1')
+	detection = forward_nsfw(detect_vision_frame, 'nsfw_1')
+	detection_score = numpy.max(numpy.amax(detection[:, 4:], axis = 1))
+	return bool(detection_score > 0.2)
+
+
+def detect_with_nsfw_2(vision_frame : VisionFrame) -> bool:
+	detect_vision_frame = prepare_detect_frame(vision_frame, 'nsfw_2')
+	detection = forward_nsfw(detect_vision_frame, 'nsfw_2')
+	detection_score = detection[0] - detection[1]
+	return bool(detection_score > 0.25)
+
+
+def detect_with_nsfw_3(vision_frame : VisionFrame) -> bool:
+	detect_vision_frame = prepare_detect_frame(vision_frame, 'nsfw_3')
+	detection = forward_nsfw(detect_vision_frame, 'nsfw_3')
+	detection_score = (detection[2] + detection[3]) - (detection[0] + detection[1])
+	return bool(detection_score > 10.5)
+
+
+def forward_nsfw(vision_frame : VisionFrame, nsfw_model : str) -> Detection:
+	content_analyser = get_inference_pool().get(nsfw_model)
+
+	with conditional_thread_semaphore():
+		detection = content_analyser.run(None,
+		{
+			'input': vision_frame
+		})[0]
+
+	if nsfw_model in [ 'nsfw_2', 'nsfw_3' ]:
+		return detection[0]
+
+	return detection
+
+
+def prepare_detect_frame(temp_vision_frame : VisionFrame, model_name : str) -> VisionFrame:
+	model_set = create_static_model_set('full').get(model_name)
+	model_size = model_set.get('size')
+	model_mean = model_set.get('mean')
+	model_standard_deviation = model_set.get('standard_deviation')
+
+	detect_vision_frame = fit_frame(temp_vision_frame, model_size)
+	detect_vision_frame = detect_vision_frame[:, :, ::-1] / 255.0
+	detect_vision_frame -= model_mean
+	detect_vision_frame /= model_standard_deviation
+	detect_vision_frame = numpy.expand_dims(detect_vision_frame.transpose(2, 0, 1), axis = 0).astype(numpy.float32)
+	return detect_vision_frame

facefusion/core.py

@@ -0,0 +1,517 @@
+import inspect
+import itertools
+import shutil
+import signal
+import sys
+from time import time
+
+import numpy
+
+from facefusion import benchmarker, cli_helper, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, hash_helper, logger, process_manager, state_manager, video_manager, voice_extractor, wording
+from facefusion.args import apply_args, collect_job_args, reduce_job_args, reduce_step_args
+from facefusion.common_helper import get_first
+from facefusion.content_analyser import analyse_image, analyse_video
+from facefusion.download import conditional_download_hashes, conditional_download_sources
+from facefusion.exit_helper import hard_exit, signal_exit
+from facefusion.face_analyser import get_average_face, get_many_faces, get_one_face
+from facefusion.face_selector import sort_and_filter_faces
+from facefusion.face_store import append_reference_face, clear_reference_faces, get_reference_faces
+from facefusion.ffmpeg import copy_image, extract_frames, finalize_image, merge_video, replace_audio, restore_audio
+from facefusion.filesystem import filter_audio_paths, get_file_name, is_image, is_video, resolve_file_paths, resolve_file_pattern
+from facefusion.jobs import job_helper, job_manager, job_runner
+from facefusion.jobs.job_list import compose_job_list
+from facefusion.memory import limit_system_memory
+from facefusion.processors.core import get_processors_modules
+from facefusion.program import create_program
+from facefusion.program_helper import validate_args
+from facefusion.temp_helper import clear_temp_directory, create_temp_directory, get_temp_file_path, move_temp_file, resolve_temp_frame_paths
+from facefusion.types import Args, ErrorCode
+from facefusion.vision import pack_resolution, read_image, read_static_images, read_video_frame, restrict_image_resolution, restrict_trim_frame, restrict_video_fps, restrict_video_resolution, unpack_resolution
+
+
+def cli() -> None:
+	if pre_check():
+		signal.signal(signal.SIGINT, signal_exit)
+		program = create_program()
+
+		if validate_args(program):
+			args = vars(program.parse_args())
+			apply_args(args, state_manager.init_item)
+
+			if state_manager.get_item('command'):
+				logger.init(state_manager.get_item('log_level'))
+				route(args)
+			else:
+				program.print_help()
+		else:
+			hard_exit(2)
+	else:
+		hard_exit(2)
+
+
+def route(args : Args) -> None:
+	system_memory_limit = state_manager.get_item('system_memory_limit')
+
+	if system_memory_limit and system_memory_limit > 0:
+		limit_system_memory(system_memory_limit)
+
+	if state_manager.get_item('command') == 'force-download':
+		error_code = force_download()
+		return hard_exit(error_code)
+
+	if state_manager.get_item('command') == 'benchmark':
+		if not common_pre_check() or not processors_pre_check() or not benchmarker.pre_check():
+			return hard_exit(2)
+		benchmarker.render()
+
+	if state_manager.get_item('command') in [ 'job-list', 'job-create', 'job-submit', 'job-submit-all', 'job-delete', 'job-delete-all', 'job-add-step', 'job-remix-step', 'job-insert-step', 'job-remove-step' ]:
+		if not job_manager.init_jobs(state_manager.get_item('jobs_path')):
+			hard_exit(1)
+		error_code = route_job_manager(args)
+		hard_exit(error_code)
+
+	if state_manager.get_item('command') == 'run':
+		import facefusion.uis.core as ui
+
+		if not common_pre_check() or not processors_pre_check():
+			return hard_exit(2)
+		for ui_layout in ui.get_ui_layouts_modules(state_manager.get_item('ui_layouts')):
+			if not ui_layout.pre_check():
+				return hard_exit(2)
+		ui.init()
+		ui.launch()
+
+	if state_manager.get_item('command') == 'headless-run':
+		if not job_manager.init_jobs(state_manager.get_item('jobs_path')):
+			hard_exit(1)
+		error_core = process_headless(args)
+		hard_exit(error_core)
+
+	if state_manager.get_item('command') == 'batch-run':
+		if not job_manager.init_jobs(state_manager.get_item('jobs_path')):
+			hard_exit(1)
+		error_core = process_batch(args)
+		hard_exit(error_core)
+
+	if state_manager.get_item('command') in [ 'job-run', 'job-run-all', 'job-retry', 'job-retry-all' ]:
+		if not job_manager.init_jobs(state_manager.get_item('jobs_path')):
+			hard_exit(1)
+		error_code = route_job_runner()
+		hard_exit(error_code)
+
+
+def pre_check() -> bool:
+	if sys.version_info < (3, 10):
+		logger.error(wording.get('python_not_supported').format(version = '3.10'), __name__)
+		return False
+
+	if not shutil.which('curl'):
+		logger.error(wording.get('curl_not_installed'), __name__)
+		return False
+
+	if not shutil.which('ffmpeg'):
+		logger.error(wording.get('ffmpeg_not_installed'), __name__)
+		return False
+	return True
+
+
+def common_pre_check() -> bool:
+	common_modules =\
+	[
+		content_analyser,
+		face_classifier,
+		face_detector,
+		face_landmarker,
+		face_masker,
+		face_recognizer,
+		voice_extractor
+	]
+
+	content_analyser_content = inspect.getsource(content_analyser).encode()
+	is_valid = hash_helper.create_hash(content_analyser_content) == 'b159fd9d'
+
+	return all(module.pre_check() for module in common_modules)
+
+
+def processors_pre_check() -> bool:
+	for processor_module in get_processors_modules(state_manager.get_item('processors')):
+		if not processor_module.pre_check():
+			return False
+	return True
+
+
+def force_download() -> ErrorCode:
+	common_modules =\
+	[
+		content_analyser,
+		face_classifier,
+		face_detector,
+		face_landmarker,
+		face_masker,
+		face_recognizer,
+		voice_extractor
+	]
+	available_processors = [ get_file_name(file_path) for file_path in resolve_file_paths('facefusion/processors/modules') ]
+	processor_modules = get_processors_modules(available_processors)
+
+	for module in common_modules + processor_modules:
+		if hasattr(module, 'create_static_model_set'):
+			for model in module.create_static_model_set(state_manager.get_item('download_scope')).values():
+				model_hash_set = model.get('hashes')
+				model_source_set = model.get('sources')
+
+				if model_hash_set and model_source_set:
+					if not conditional_download_hashes(model_hash_set) or not conditional_download_sources(model_source_set):
+						return 1
+
+	return 0
+
+
+def route_job_manager(args : Args) -> ErrorCode:
+	if state_manager.get_item('command') == 'job-list':
+		job_headers, job_contents = compose_job_list(state_manager.get_item('job_status'))
+
+		if job_contents:
+			cli_helper.render_table(job_headers, job_contents)
+			return 0
+		return 1
+
+	if state_manager.get_item('command') == 'job-create':
+		if job_manager.create_job(state_manager.get_item('job_id')):
+			logger.info(wording.get('job_created').format(job_id = state_manager.get_item('job_id')), __name__)
+			return 0
+		logger.error(wording.get('job_not_created').format(job_id = state_manager.get_item('job_id')), __name__)
+		return 1
+
+	if state_manager.get_item('command') == 'job-submit':
+		if job_manager.submit_job(state_manager.get_item('job_id')):
+			logger.info(wording.get('job_submitted').format(job_id = state_manager.get_item('job_id')), __name__)
+			return 0
+		logger.error(wording.get('job_not_submitted').format(job_id = state_manager.get_item('job_id')), __name__)
+		return 1
+
+	if state_manager.get_item('command') == 'job-submit-all':
+		if job_manager.submit_jobs(state_manager.get_item('halt_on_error')):
+			logger.info(wording.get('job_all_submitted'), __name__)
+			return 0
+		logger.error(wording.get('job_all_not_submitted'), __name__)
+		return 1
+
+	if state_manager.get_item('command') == 'job-delete':
+		if job_manager.delete_job(state_manager.get_item('job_id')):
+			logger.info(wording.get('job_deleted').format(job_id = state_manager.get_item('job_id')), __name__)
+			return 0
+		logger.error(wording.get('job_not_deleted').format(job_id = state_manager.get_item('job_id')), __name__)
+		return 1
+
+	if state_manager.get_item('command') == 'job-delete-all':
+		if job_manager.delete_jobs(state_manager.get_item('halt_on_error')):
+			logger.info(wording.get('job_all_deleted'), __name__)
+			return 0
+		logger.error(wording.get('job_all_not_deleted'), __name__)
+		return 1
+
+	if state_manager.get_item('command') == 'job-add-step':
+		step_args = reduce_step_args(args)
+
+		if job_manager.add_step(state_manager.get_item('job_id'), step_args):
+			logger.info(wording.get('job_step_added').format(job_id = state_manager.get_item('job_id')), __name__)
+			return 0
+		logger.error(wording.get('job_step_not_added').format(job_id = state_manager.get_item('job_id')), __name__)
+		return 1
+
+	if state_manager.get_item('command') == 'job-remix-step':
+		step_args = reduce_step_args(args)
+
+		if job_manager.remix_step(state_manager.get_item('job_id'), state_manager.get_item('step_index'), step_args):
+			logger.info(wording.get('job_remix_step_added').format(job_id = state_manager.get_item('job_id'), step_index = state_manager.get_item('step_index')), __name__)
+			return 0
+		logger.error(wording.get('job_remix_step_not_added').format(job_id = state_manager.get_item('job_id'), step_index = state_manager.get_item('step_index')), __name__)
+		return 1
+
+	if state_manager.get_item('command') == 'job-insert-step':
+		step_args = reduce_step_args(args)
+
+		if job_manager.insert_step(state_manager.get_item('job_id'), state_manager.get_item('step_index'), step_args):
+			logger.info(wording.get('job_step_inserted').format(job_id = state_manager.get_item('job_id'), step_index = state_manager.get_item('step_index')), __name__)
+			return 0
+		logger.error(wording.get('job_step_not_inserted').format(job_id = state_manager.get_item('job_id'), step_index = state_manager.get_item('step_index')), __name__)
+		return 1
+
+	if state_manager.get_item('command') == 'job-remove-step':
+		if job_manager.remove_step(state_manager.get_item('job_id'), state_manager.get_item('step_index')):
+			logger.info(wording.get('job_step_removed').format(job_id = state_manager.get_item('job_id'), step_index = state_manager.get_item('step_index')), __name__)
+			return 0
+		logger.error(wording.get('job_step_not_removed').format(job_id = state_manager.get_item('job_id'), step_index = state_manager.get_item('step_index')), __name__)
+		return 1
+	return 1
+
+
+def route_job_runner() -> ErrorCode:
+	if state_manager.get_item('command') == 'job-run':
+		logger.info(wording.get('running_job').format(job_id = state_manager.get_item('job_id')), __name__)
+		if job_runner.run_job(state_manager.get_item('job_id'), process_step):
+			logger.info(wording.get('processing_job_succeed').format(job_id = state_manager.get_item('job_id')), __name__)
+			return 0
+		logger.info(wording.get('processing_job_failed').format(job_id = state_manager.get_item('job_id')), __name__)
+		return 1
+
+	if state_manager.get_item('command') == 'job-run-all':
+		logger.info(wording.get('running_jobs'), __name__)
+		if job_runner.run_jobs(process_step, state_manager.get_item('halt_on_error')):
+			logger.info(wording.get('processing_jobs_succeed'), __name__)
+			return 0
+		logger.info(wording.get('processing_jobs_failed'), __name__)
+		return 1
+
+	if state_manager.get_item('command') == 'job-retry':
+		logger.info(wording.get('retrying_job').format(job_id = state_manager.get_item('job_id')), __name__)
+		if job_runner.retry_job(state_manager.get_item('job_id'), process_step):
+			logger.info(wording.get('processing_job_succeed').format(job_id = state_manager.get_item('job_id')), __name__)
+			return 0
+		logger.info(wording.get('processing_job_failed').format(job_id = state_manager.get_item('job_id')), __name__)
+		return 1
+
+	if state_manager.get_item('command') == 'job-retry-all':
+		logger.info(wording.get('retrying_jobs'), __name__)
+		if job_runner.retry_jobs(process_step, state_manager.get_item('halt_on_error')):
+			logger.info(wording.get('processing_jobs_succeed'), __name__)
+			return 0
+		logger.info(wording.get('processing_jobs_failed'), __name__)
+		return 1
+	return 2
+
+
+def process_headless(args : Args) -> ErrorCode:
+	job_id = job_helper.suggest_job_id('headless')
+	step_args = reduce_step_args(args)
+
+	if job_manager.create_job(job_id) and job_manager.add_step(job_id, step_args) and job_manager.submit_job(job_id) and job_runner.run_job(job_id, process_step):
+		return 0
+	return 1
+
+
+def process_batch(args : Args) -> ErrorCode:
+	job_id = job_helper.suggest_job_id('batch')
+	step_args = reduce_step_args(args)
+	job_args = reduce_job_args(args)
+	source_paths = resolve_file_pattern(job_args.get('source_pattern'))
+	target_paths = resolve_file_pattern(job_args.get('target_pattern'))
+
+	if job_manager.create_job(job_id):
+		if source_paths and target_paths:
+			for index, (source_path, target_path) in enumerate(itertools.product(source_paths, target_paths)):
+				step_args['source_paths'] = [ source_path ]
+				step_args['target_path'] = target_path
+				step_args['output_path'] = job_args.get('output_pattern').format(index = index)
+				if not job_manager.add_step(job_id, step_args):
+					return 1
+			if job_manager.submit_job(job_id) and job_runner.run_job(job_id, process_step):
+				return 0
+
+		if not source_paths and target_paths:
+			for index, target_path in enumerate(target_paths):
+				step_args['target_path'] = target_path
+				step_args['output_path'] = job_args.get('output_pattern').format(index = index)
+				if not job_manager.add_step(job_id, step_args):
+					return 1
+			if job_manager.submit_job(job_id) and job_runner.run_job(job_id, process_step):
+				return 0
+	return 1
+
+
+def process_step(job_id : str, step_index : int, step_args : Args) -> bool:
+	clear_reference_faces()
+	step_total = job_manager.count_step_total(job_id)
+	step_args.update(collect_job_args())
+	apply_args(step_args, state_manager.set_item)
+
+	logger.info(wording.get('processing_step').format(step_current = step_index + 1, step_total = step_total), __name__)
+	if common_pre_check() and processors_pre_check():
+		error_code = conditional_process()
+		return error_code == 0
+	return False
+
+
+def conditional_process() -> ErrorCode:
+	start_time = time()
+
+	for processor_module in get_processors_modules(state_manager.get_item('processors')):
+		if not processor_module.pre_process('output'):
+			return 2
+
+	conditional_append_reference_faces()
+
+	if is_image(state_manager.get_item('target_path')):
+		return process_image(start_time)
+	if is_video(state_manager.get_item('target_path')):
+		return process_video(start_time)
+
+	return 0
+
+
+def conditional_append_reference_faces() -> None:
+	if 'reference' in state_manager.get_item('face_selector_mode') and not get_reference_faces():
+		source_frames = read_static_images(state_manager.get_item('source_paths'))
+		source_faces = get_many_faces(source_frames)
+		source_face = get_average_face(source_faces)
+		if is_video(state_manager.get_item('target_path')):
+			reference_frame = read_video_frame(state_manager.get_item('target_path'), state_manager.get_item('reference_frame_number'))
+		else:
+			reference_frame = read_image(state_manager.get_item('target_path'))
+		reference_faces = sort_and_filter_faces(get_many_faces([ reference_frame ]))
+		reference_face = get_one_face(reference_faces, state_manager.get_item('reference_face_position'))
+		append_reference_face('origin', reference_face)
+
+		if source_face and reference_face:
+			for processor_module in get_processors_modules(state_manager.get_item('processors')):
+				abstract_reference_frame = processor_module.get_reference_frame(source_face, reference_face, reference_frame)
+				if numpy.any(abstract_reference_frame):
+					abstract_reference_faces = sort_and_filter_faces(get_many_faces([ abstract_reference_frame ]))
+					abstract_reference_face = get_one_face(abstract_reference_faces, state_manager.get_item('reference_face_position'))
+					append_reference_face(processor_module.__name__, abstract_reference_face)
+
+
+def process_image(start_time : float) -> ErrorCode:
+	if analyse_image(state_manager.get_item('target_path')):
+		return 3
+
+	logger.debug(wording.get('clearing_temp'), __name__)
+	clear_temp_directory(state_manager.get_item('target_path'))
+	logger.debug(wording.get('creating_temp'), __name__)
+	create_temp_directory(state_manager.get_item('target_path'))
+
+	process_manager.start()
+	temp_image_resolution = pack_resolution(restrict_image_resolution(state_manager.get_item('target_path'), unpack_resolution(state_manager.get_item('output_image_resolution'))))
+	logger.info(wording.get('copying_image').format(resolution = temp_image_resolution), __name__)
+	if copy_image(state_manager.get_item('target_path'), temp_image_resolution):
+		logger.debug(wording.get('copying_image_succeed'), __name__)
+	else:
+		logger.error(wording.get('copying_image_failed'), __name__)
+		process_manager.end()
+		return 1
+
+	temp_image_path = get_temp_file_path(state_manager.get_item('target_path'))
+	for processor_module in get_processors_modules(state_manager.get_item('processors')):
+		logger.info(wording.get('processing'), processor_module.__name__)
+		processor_module.process_image(state_manager.get_item('source_paths'), temp_image_path, temp_image_path)
+		processor_module.post_process()
+	if is_process_stopping():
+		process_manager.end()
+		return 4
+
+	logger.info(wording.get('finalizing_image').format(resolution = state_manager.get_item('output_image_resolution')), __name__)
+	if finalize_image(state_manager.get_item('target_path'), state_manager.get_item('output_path'), state_manager.get_item('output_image_resolution')):
+		logger.debug(wording.get('finalizing_image_succeed'), __name__)
+	else:
+		logger.warn(wording.get('finalizing_image_skipped'), __name__)
+
+	logger.debug(wording.get('clearing_temp'), __name__)
+	clear_temp_directory(state_manager.get_item('target_path'))
+
+	if is_image(state_manager.get_item('output_path')):
+		seconds = '{:.2f}'.format((time() - start_time) % 60)
+		logger.info(wording.get('processing_image_succeed').format(seconds = seconds), __name__)
+	else:
+		logger.error(wording.get('processing_image_failed'), __name__)
+		process_manager.end()
+		return 1
+	process_manager.end()
+	return 0
+
+
+def process_video(start_time : float) -> ErrorCode:
+	trim_frame_start, trim_frame_end = restrict_trim_frame(state_manager.get_item('target_path'), state_manager.get_item('trim_frame_start'), state_manager.get_item('trim_frame_end'))
+	if analyse_video(state_manager.get_item('target_path'), trim_frame_start, trim_frame_end):
+		return 3
+
+	logger.debug(wording.get('clearing_temp'), __name__)
+	clear_temp_directory(state_manager.get_item('target_path'))
+	logger.debug(wording.get('creating_temp'), __name__)
+	create_temp_directory(state_manager.get_item('target_path'))
+
+	process_manager.start()
+	temp_video_resolution = pack_resolution(restrict_video_resolution(state_manager.get_item('target_path'), unpack_resolution(state_manager.get_item('output_video_resolution'))))
+	temp_video_fps = restrict_video_fps(state_manager.get_item('target_path'), state_manager.get_item('output_video_fps'))
+	logger.info(wording.get('extracting_frames').format(resolution = temp_video_resolution, fps = temp_video_fps), __name__)
+	if extract_frames(state_manager.get_item('target_path'), temp_video_resolution, temp_video_fps, trim_frame_start, trim_frame_end):
+		logger.debug(wording.get('extracting_frames_succeed'), __name__)
+	else:
+		if is_process_stopping():
+			process_manager.end()
+			return 4
+		logger.error(wording.get('extracting_frames_failed'), __name__)
+		process_manager.end()
+		return 1
+
+	temp_frame_paths = resolve_temp_frame_paths(state_manager.get_item('target_path'))
+	if temp_frame_paths:
+		for processor_module in get_processors_modules(state_manager.get_item('processors')):
+			logger.info(wording.get('processing'), processor_module.__name__)
+			processor_module.process_video(state_manager.get_item('source_paths'), temp_frame_paths)
+			processor_module.post_process()
+		if is_process_stopping():
+			return 4
+	else:
+		logger.error(wording.get('temp_frames_not_found'), __name__)
+		process_manager.end()
+		return 1
+
+	logger.info(wording.get('merging_video').format(resolution = state_manager.get_item('output_video_resolution'), fps = state_manager.get_item('output_video_fps')), __name__)
+	if merge_video(state_manager.get_item('target_path'), temp_video_fps, state_manager.get_item('output_video_resolution'), state_manager.get_item('output_video_fps'), trim_frame_start, trim_frame_end):
+		logger.debug(wording.get('merging_video_succeed'), __name__)
+	else:
+		if is_process_stopping():
+			process_manager.end()
+			return 4
+		logger.error(wording.get('merging_video_failed'), __name__)
+		process_manager.end()
+		return 1
+
+	if state_manager.get_item('output_audio_volume') == 0:
+		logger.info(wording.get('skipping_audio'), __name__)
+		move_temp_file(state_manager.get_item('target_path'), state_manager.get_item('output_path'))
+	else:
+		source_audio_path = get_first(filter_audio_paths(state_manager.get_item('source_paths')))
+		if source_audio_path:
+			if replace_audio(state_manager.get_item('target_path'), source_audio_path, state_manager.get_item('output_path')):
+				video_manager.clear_video_pool()
+				logger.debug(wording.get('replacing_audio_succeed'), __name__)
+			else:
+				video_manager.clear_video_pool()
+				if is_process_stopping():
+					process_manager.end()
+					return 4
+				logger.warn(wording.get('replacing_audio_skipped'), __name__)
+				move_temp_file(state_manager.get_item('target_path'), state_manager.get_item('output_path'))
+		else:
+			if restore_audio(state_manager.get_item('target_path'), state_manager.get_item('output_path'), trim_frame_start, trim_frame_end):
+				video_manager.clear_video_pool()
+				logger.debug(wording.get('restoring_audio_succeed'), __name__)
+			else:
+				video_manager.clear_video_pool()
+				if is_process_stopping():
+					process_manager.end()
+					return 4
+				logger.warn(wording.get('restoring_audio_skipped'), __name__)
+				move_temp_file(state_manager.get_item('target_path'), state_manager.get_item('output_path'))
+
+	logger.debug(wording.get('clearing_temp'), __name__)
+	clear_temp_directory(state_manager.get_item('target_path'))
+
+	if is_video(state_manager.get_item('output_path')):
+		seconds = '{:.2f}'.format((time() - start_time))
+		logger.info(wording.get('processing_video_succeed').format(seconds = seconds), __name__)
+	else:
+		logger.error(wording.get('processing_video_failed'), __name__)
+		process_manager.end()
+		return 1
+	process_manager.end()
+	return 0
+
+
+def is_process_stopping() -> bool:
+	if process_manager.is_stopping():
+		process_manager.end()
+		logger.info(wording.get('processing_stopped'), __name__)
+	return process_manager.is_pending()

facefusion/curl_builder.py

@@ -0,0 +1,27 @@
+import itertools
+import shutil
+
+from facefusion import metadata
+from facefusion.types import Commands
+
+
+def run(commands : Commands) -> Commands:
+	user_agent = metadata.get('name') + '/' + metadata.get('version')
+
+	return [ shutil.which('curl'), '--user-agent', user_agent, '--insecure', '--location', '--silent' ] + commands
+
+
+def chain(*commands : Commands) -> Commands:
+	return list(itertools.chain(*commands))
+
+
+def head(url : str) -> Commands:
+	return [ '-I', url ]
+
+
+def download(url : str, download_file_path : str) -> Commands:
+	return [ '--create-dirs', '--continue-at', '-', '--output', download_file_path, url ]
+
+
+def set_timeout(timeout : int) -> Commands:
+	return [ '--connect-timeout', str(timeout) ]

facefusion/date_helper.py

@@ -0,0 +1,28 @@
+from datetime import datetime, timedelta
+from typing import Optional, Tuple
+
+from facefusion import wording
+
+
+def get_current_date_time() -> datetime:
+	return datetime.now().astimezone()
+
+
+def split_time_delta(time_delta : timedelta) -> Tuple[int, int, int, int]:
+	days, hours = divmod(time_delta.total_seconds(), 86400)
+	hours, minutes = divmod(hours, 3600)
+	minutes, seconds = divmod(minutes, 60)
+	return int(days), int(hours), int(minutes), int(seconds)
+
+
+def describe_time_ago(date_time : datetime) -> Optional[str]:
+	time_ago = datetime.now(date_time.tzinfo) - date_time
+	days, hours, minutes, _ = split_time_delta(time_ago)
+
+	if timedelta(days = 1) < time_ago:
+		return wording.get('time_ago_days').format(days = days, hours = hours, minutes = minutes)
+	if timedelta(hours = 1) < time_ago:
+		return wording.get('time_ago_hours').format(hours = hours, minutes = minutes)
+	if timedelta(minutes = 1) < time_ago:
+		return wording.get('time_ago_minutes').format(minutes = minutes)
+	return wording.get('time_ago_now')

facefusion/download.py

@@ -0,0 +1,174 @@
+import os
+import subprocess
+from functools import lru_cache
+from typing import List, Optional, Tuple
+from urllib.parse import urlparse
+
+from tqdm import tqdm
+
+import facefusion.choices
+from facefusion import curl_builder, logger, process_manager, state_manager, wording
+from facefusion.filesystem import get_file_name, get_file_size, is_file, remove_file
+from facefusion.hash_helper import validate_hash
+from facefusion.types import Commands, DownloadProvider, DownloadSet
+
+
+def open_curl(commands : Commands) -> subprocess.Popen[bytes]:
+	commands = curl_builder.run(commands)
+	return subprocess.Popen(commands, stdin = subprocess.PIPE, stdout = subprocess.PIPE)
+
+
+def conditional_download(download_directory_path : str, urls : List[str]) -> None:
+	for url in urls:
+		download_file_name = os.path.basename(urlparse(url).path)
+		download_file_path = os.path.join(download_directory_path, download_file_name)
+		initial_size = get_file_size(download_file_path)
+		download_size = get_static_download_size(url)
+
+		if initial_size < download_size:
+			with tqdm(total = download_size, initial = initial_size, desc = wording.get('downloading'), unit = 'B', unit_scale = True, unit_divisor = 1024, ascii = ' =', disable = state_manager.get_item('log_level') in [ 'warn', 'error' ]) as progress:
+				commands = curl_builder.chain(
+					curl_builder.download(url, download_file_path),
+					curl_builder.set_timeout(10)
+				)
+				open_curl(commands)
+				current_size = initial_size
+				progress.set_postfix(download_providers = state_manager.get_item('download_providers'), file_name = download_file_name)
+
+				while current_size < download_size:
+					if is_file(download_file_path):
+						current_size = get_file_size(download_file_path)
+						progress.update(current_size - progress.n)
+
+
+@lru_cache(maxsize = None)
+def get_static_download_size(url : str) -> int:
+	commands = curl_builder.chain(
+		curl_builder.head(url),
+		curl_builder.set_timeout(5)
+	)
+	process = open_curl(commands)
+	lines = reversed(process.stdout.readlines())
+
+	for line in lines:
+		__line__ = line.decode().lower()
+		if 'content-length:' in __line__:
+			_, content_length = __line__.split('content-length:')
+			return int(content_length)
+
+	return 0
+
+
+@lru_cache(maxsize = None)
+def ping_static_url(url : str) -> bool:
+	commands = curl_builder.chain(
+		curl_builder.head(url),
+		curl_builder.set_timeout(5)
+	)
+	process = open_curl(commands)
+	process.communicate()
+	return process.returncode == 0
+
+
+def conditional_download_hashes(hash_set : DownloadSet) -> bool:
+	hash_paths = [ hash_set.get(hash_key).get('path') for hash_key in hash_set.keys() ]
+
+	process_manager.check()
+	_, invalid_hash_paths = validate_hash_paths(hash_paths)
+	if invalid_hash_paths:
+		for index in hash_set:
+			if hash_set.get(index).get('path') in invalid_hash_paths:
+				invalid_hash_url = hash_set.get(index).get('url')
+				if invalid_hash_url:
+					download_directory_path = os.path.dirname(hash_set.get(index).get('path'))
+					conditional_download(download_directory_path, [ invalid_hash_url ])
+
+	valid_hash_paths, invalid_hash_paths = validate_hash_paths(hash_paths)
+
+	for valid_hash_path in valid_hash_paths:
+		valid_hash_file_name = get_file_name(valid_hash_path)
+		logger.debug(wording.get('validating_hash_succeed').format(hash_file_name = valid_hash_file_name), __name__)
+	for invalid_hash_path in invalid_hash_paths:
+		invalid_hash_file_name = get_file_name(invalid_hash_path)
+		logger.error(wording.get('validating_hash_failed').format(hash_file_name = invalid_hash_file_name), __name__)
+
+	if not invalid_hash_paths:
+		process_manager.end()
+	return not invalid_hash_paths
+
+
+def conditional_download_sources(source_set : DownloadSet) -> bool:
+	source_paths = [ source_set.get(source_key).get('path') for source_key in source_set.keys() ]
+
+	process_manager.check()
+	_, invalid_source_paths = validate_source_paths(source_paths)
+	if invalid_source_paths:
+		for index in source_set:
+			if source_set.get(index).get('path') in invalid_source_paths:
+				invalid_source_url = source_set.get(index).get('url')
+				if invalid_source_url:
+					download_directory_path = os.path.dirname(source_set.get(index).get('path'))
+					conditional_download(download_directory_path, [ invalid_source_url ])
+
+	valid_source_paths, invalid_source_paths = validate_source_paths(source_paths)
+
+	for valid_source_path in valid_source_paths:
+		valid_source_file_name = get_file_name(valid_source_path)
+		logger.debug(wording.get('validating_source_succeed').format(source_file_name = valid_source_file_name), __name__)
+	for invalid_source_path in invalid_source_paths:
+		invalid_source_file_name = get_file_name(invalid_source_path)
+		logger.error(wording.get('validating_source_failed').format(source_file_name = invalid_source_file_name), __name__)
+
+		if remove_file(invalid_source_path):
+			logger.error(wording.get('deleting_corrupt_source').format(source_file_name = invalid_source_file_name), __name__)
+
+	if not invalid_source_paths:
+		process_manager.end()
+	return not invalid_source_paths
+
+
+def validate_hash_paths(hash_paths : List[str]) -> Tuple[List[str], List[str]]:
+	valid_hash_paths = []
+	invalid_hash_paths = []
+
+	for hash_path in hash_paths:
+		if is_file(hash_path):
+			valid_hash_paths.append(hash_path)
+		else:
+			invalid_hash_paths.append(hash_path)
+
+	return valid_hash_paths, invalid_hash_paths
+
+
+def validate_source_paths(source_paths : List[str]) -> Tuple[List[str], List[str]]:
+	valid_source_paths = []
+	invalid_source_paths = []
+
+	for source_path in source_paths:
+		if validate_hash(source_path):
+			valid_source_paths.append(source_path)
+		else:
+			invalid_source_paths.append(source_path)
+
+	return valid_source_paths, invalid_source_paths
+
+
+def resolve_download_url(base_name : str, file_name : str) -> Optional[str]:
+	download_providers = state_manager.get_item('download_providers')
+
+	for download_provider in download_providers:
+		download_url = resolve_download_url_by_provider(download_provider, base_name, file_name)
+		if download_url:
+			return download_url
+
+	return None
+
+
+def resolve_download_url_by_provider(download_provider : DownloadProvider, base_name : str, file_name : str) -> Optional[str]:
+	download_provider_value = facefusion.choices.download_provider_set.get(download_provider)
+
+	for download_provider_url in download_provider_value.get('urls'):
+		if ping_static_url(download_provider_url):
+			return download_provider_url + download_provider_value.get('path').format(base_name = base_name, file_name = file_name)
+
+	return None

facefusion/execution.py

@@ -0,0 +1,156 @@
+import shutil
+import subprocess
+import xml.etree.ElementTree as ElementTree
+from functools import lru_cache
+from typing import List, Optional
+
+from onnxruntime import get_available_providers, set_default_logger_severity
+
+import facefusion.choices
+from facefusion.types import ExecutionDevice, ExecutionProvider, InferenceSessionProvider, ValueAndUnit
+
+set_default_logger_severity(3)
+
+
+def has_execution_provider(execution_provider : ExecutionProvider) -> bool:
+	return execution_provider in get_available_execution_providers()
+
+
+def get_available_execution_providers() -> List[ExecutionProvider]:
+	inference_session_providers = get_available_providers()
+	available_execution_providers : List[ExecutionProvider] = []
+
+	for execution_provider, execution_provider_value in facefusion.choices.execution_provider_set.items():
+		if execution_provider_value in inference_session_providers:
+			index = facefusion.choices.execution_providers.index(execution_provider)
+			available_execution_providers.insert(index, execution_provider)
+
+	return available_execution_providers
+
+
+def create_inference_session_providers(execution_device_id : str, execution_providers : List[ExecutionProvider]) -> List[InferenceSessionProvider]:
+	inference_session_providers : List[InferenceSessionProvider] = []
+
+	for execution_provider in execution_providers:
+		if execution_provider == 'cuda':
+			inference_session_providers.append((facefusion.choices.execution_provider_set.get(execution_provider),
+			{
+				'device_id': execution_device_id,
+				'cudnn_conv_algo_search': resolve_cudnn_conv_algo_search()
+			}))
+		if execution_provider == 'tensorrt':
+			inference_session_providers.append((facefusion.choices.execution_provider_set.get(execution_provider),
+			{
+				'device_id': execution_device_id,
+				'trt_engine_cache_enable': True,
+				'trt_engine_cache_path': '.caches',
+				'trt_timing_cache_enable': True,
+				'trt_timing_cache_path': '.caches',
+				'trt_builder_optimization_level': 5
+			}))
+		if execution_provider in [ 'directml', 'rocm' ]:
+			inference_session_providers.append((facefusion.choices.execution_provider_set.get(execution_provider),
+			{
+				'device_id': execution_device_id
+			}))
+		if execution_provider == 'openvino':
+			inference_session_providers.append((facefusion.choices.execution_provider_set.get(execution_provider),
+			{
+				'device_type': resolve_openvino_device_type(execution_device_id),
+				'precision': 'FP32'
+			}))
+		if execution_provider == 'coreml':
+			inference_session_providers.append((facefusion.choices.execution_provider_set.get(execution_provider),
+			{
+				'SpecializationStrategy': 'FastPrediction',
+				'ModelCacheDirectory': '.caches'
+			}))
+
+	if 'cpu' in execution_providers:
+		inference_session_providers.append(facefusion.choices.execution_provider_set.get('cpu'))
+
+	return inference_session_providers
+
+
+def resolve_cudnn_conv_algo_search() -> str:
+	execution_devices = detect_static_execution_devices()
+	product_names = ('GeForce GTX 1630', 'GeForce GTX 1650', 'GeForce GTX 1660')
+
+	for execution_device in execution_devices:
+		if execution_device.get('product').get('name').startswith(product_names):
+			return 'DEFAULT'
+
+	return 'EXHAUSTIVE'
+
+
+def resolve_openvino_device_type(execution_device_id : str) -> str:
+	if execution_device_id == '0':
+		return 'GPU'
+	if execution_device_id == '∞':
+		return 'MULTI:GPU'
+	return 'GPU.' + execution_device_id
+
+
+def run_nvidia_smi() -> subprocess.Popen[bytes]:
+	commands = [ shutil.which('nvidia-smi'), '--query', '--xml-format' ]
+	return subprocess.Popen(commands, stdout = subprocess.PIPE)
+
+
+@lru_cache(maxsize = None)
+def detect_static_execution_devices() -> List[ExecutionDevice]:
+	return detect_execution_devices()
+
+
+def detect_execution_devices() -> List[ExecutionDevice]:
+	execution_devices : List[ExecutionDevice] = []
+
+	try:
+		output, _ = run_nvidia_smi().communicate()
+		root_element = ElementTree.fromstring(output)
+	except Exception:
+		root_element = ElementTree.Element('xml')
+
+	for gpu_element in root_element.findall('gpu'):
+		execution_devices.append(
+		{
+			'driver_version': root_element.findtext('driver_version'),
+			'framework':
+			{
+				'name': 'CUDA',
+				'version': root_element.findtext('cuda_version')
+			},
+			'product':
+			{
+				'vendor': 'NVIDIA',
+				'name': gpu_element.findtext('product_name').replace('NVIDIA', '').strip()
+			},
+			'video_memory':
+			{
+				'total': create_value_and_unit(gpu_element.findtext('fb_memory_usage/total')),
+				'free': create_value_and_unit(gpu_element.findtext('fb_memory_usage/free'))
+			},
+			'temperature':
+			{
+				'gpu': create_value_and_unit(gpu_element.findtext('temperature/gpu_temp')),
+				'memory': create_value_and_unit(gpu_element.findtext('temperature/memory_temp'))
+			},
+			'utilization':
+			{
+				'gpu': create_value_and_unit(gpu_element.findtext('utilization/gpu_util')),
+				'memory': create_value_and_unit(gpu_element.findtext('utilization/memory_util'))
+			}
+		})
+
+	return execution_devices
+
+
+def create_value_and_unit(text : str) -> Optional[ValueAndUnit]:
+	if ' ' in text:
+		value, unit = text.split()
+
+		return\
+		{
+			'value': int(value),
+			'unit': str(unit)
+		}
+	return None

facefusion/exit_helper.py

@@ -0,0 +1,26 @@
+import signal
+import sys
+from time import sleep
+from types import FrameType
+
+from facefusion import process_manager, state_manager
+from facefusion.temp_helper import clear_temp_directory
+from facefusion.types import ErrorCode
+
+
+def hard_exit(error_code : ErrorCode) -> None:
+	signal.signal(signal.SIGINT, signal.SIG_IGN)
+	sys.exit(error_code)
+
+
+def signal_exit(signum : int, frame : FrameType) -> None:
+	graceful_exit(0)
+
+
+def graceful_exit(error_code : ErrorCode) -> None:
+	process_manager.stop()
+	while process_manager.is_processing():
+		sleep(0.5)
+	if state_manager.get_item('target_path'):
+		clear_temp_directory(state_manager.get_item('target_path'))
+	hard_exit(error_code)

facefusion/face_analyser.py

@@ -0,0 +1,124 @@
+from typing import List, Optional
+
+import numpy
+
+from facefusion import state_manager
+from facefusion.common_helper import get_first
+from facefusion.face_classifier import classify_face
+from facefusion.face_detector import detect_faces, detect_rotated_faces
+from facefusion.face_helper import apply_nms, convert_to_face_landmark_5, estimate_face_angle, get_nms_threshold
+from facefusion.face_landmarker import detect_face_landmark, estimate_face_landmark_68_5
+from facefusion.face_recognizer import calc_embedding
+from facefusion.face_store import get_static_faces, set_static_faces
+from facefusion.types import BoundingBox, Face, FaceLandmark5, FaceLandmarkSet, FaceScoreSet, Score, VisionFrame
+
+
+def create_faces(vision_frame : VisionFrame, bounding_boxes : List[BoundingBox], face_scores : List[Score], face_landmarks_5 : List[FaceLandmark5]) -> List[Face]:
+	faces = []
+	nms_threshold = get_nms_threshold(state_manager.get_item('face_detector_model'), state_manager.get_item('face_detector_angles'))
+	keep_indices = apply_nms(bounding_boxes, face_scores, state_manager.get_item('face_detector_score'), nms_threshold)
+
+	for index in keep_indices:
+		bounding_box = bounding_boxes[index]
+		face_score = face_scores[index]
+		face_landmark_5 = face_landmarks_5[index]
+		face_landmark_5_68 = face_landmark_5
+		face_landmark_68_5 = estimate_face_landmark_68_5(face_landmark_5_68)
+		face_landmark_68 = face_landmark_68_5
+		face_landmark_score_68 = 0.0
+		face_angle = estimate_face_angle(face_landmark_68_5)
+
+		if state_manager.get_item('face_landmarker_score') > 0:
+			face_landmark_68, face_landmark_score_68 = detect_face_landmark(vision_frame, bounding_box, face_angle)
+		if face_landmark_score_68 > state_manager.get_item('face_landmarker_score'):
+			face_landmark_5_68 = convert_to_face_landmark_5(face_landmark_68)
+
+		face_landmark_set : FaceLandmarkSet =\
+		{
+			'5': face_landmark_5,
+			'5/68': face_landmark_5_68,
+			'68': face_landmark_68,
+			'68/5': face_landmark_68_5
+		}
+		face_score_set : FaceScoreSet =\
+		{
+			'detector': face_score,
+			'landmarker': face_landmark_score_68
+		}
+		embedding, normed_embedding = calc_embedding(vision_frame, face_landmark_set.get('5/68'))
+		gender, age, race = classify_face(vision_frame, face_landmark_set.get('5/68'))
+		faces.append(Face(
+			bounding_box = bounding_box,
+			score_set = face_score_set,
+			landmark_set = face_landmark_set,
+			angle = face_angle,
+			embedding = embedding,
+			normed_embedding = normed_embedding,
+			gender = gender,
+			age = age,
+			race = race
+		))
+	return faces
+
+
+def get_one_face(faces : List[Face], position : int = 0) -> Optional[Face]:
+	if faces:
+		position = min(position, len(faces) - 1)
+		return faces[position]
+	return None
+
+
+def get_average_face(faces : List[Face]) -> Optional[Face]:
+	embeddings = []
+	normed_embeddings = []
+
+	if faces:
+		first_face = get_first(faces)
+
+		for face in faces:
+			embeddings.append(face.embedding)
+			normed_embeddings.append(face.normed_embedding)
+
+		return Face(
+			bounding_box = first_face.bounding_box,
+			score_set = first_face.score_set,
+			landmark_set = first_face.landmark_set,
+			angle = first_face.angle,
+			embedding = numpy.mean(embeddings, axis = 0),
+			normed_embedding = numpy.mean(normed_embeddings, axis = 0),
+			gender = first_face.gender,
+			age = first_face.age,
+			race = first_face.race
+		)
+	return None
+
+
+def get_many_faces(vision_frames : List[VisionFrame]) -> List[Face]:
+	many_faces : List[Face] = []
+
+	for vision_frame in vision_frames:
+		if numpy.any(vision_frame):
+			static_faces = get_static_faces(vision_frame)
+			if static_faces:
+				many_faces.extend(static_faces)
+			else:
+				all_bounding_boxes = []
+				all_face_scores = []
+				all_face_landmarks_5 = []
+
+				for face_detector_angle in state_manager.get_item('face_detector_angles'):
+					if face_detector_angle == 0:
+						bounding_boxes, face_scores, face_landmarks_5 = detect_faces(vision_frame)
+					else:
+						bounding_boxes, face_scores, face_landmarks_5 = detect_rotated_faces(vision_frame, face_detector_angle)
+					all_bounding_boxes.extend(bounding_boxes)
+					all_face_scores.extend(face_scores)
+					all_face_landmarks_5.extend(face_landmarks_5)
+
+				if all_bounding_boxes and all_face_scores and all_face_landmarks_5 and state_manager.get_item('face_detector_score') > 0:
+					faces = create_faces(vision_frame, all_bounding_boxes, all_face_scores, all_face_landmarks_5)
+
+					if faces:
+						many_faces.extend(faces)
+						set_static_faces(vision_frame, faces)
+	return many_faces

facefusion/face_classifier.py

@@ -0,0 +1,134 @@
+from functools import lru_cache
+from typing import List, Tuple
+
+import numpy
+
+from facefusion import inference_manager
+from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
+from facefusion.face_helper import warp_face_by_face_landmark_5
+from facefusion.filesystem import resolve_relative_path
+from facefusion.thread_helper import conditional_thread_semaphore
+from facefusion.types import Age, DownloadScope, FaceLandmark5, Gender, InferencePool, ModelOptions, ModelSet, Race, VisionFrame
+
+
+@lru_cache(maxsize = None)
+def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
+	return\
+	{
+		'fairface':
+		{
+			'hashes':
+			{
+				'face_classifier':
+				{
+					'url': resolve_download_url('models-3.0.0', 'fairface.hash'),
+					'path': resolve_relative_path('../.assets/models/fairface.hash')
+				}
+			},
+			'sources':
+			{
+				'face_classifier':
+				{
+					'url': resolve_download_url('models-3.0.0', 'fairface.onnx'),
+					'path': resolve_relative_path('../.assets/models/fairface.onnx')
+				}
+			},
+			'template': 'arcface_112_v2',
+			'size': (224, 224),
+			'mean': [ 0.485, 0.456, 0.406 ],
+			'standard_deviation': [ 0.229, 0.224, 0.225 ]
+		}
+	}
+
+
+def get_inference_pool() -> InferencePool:
+	model_names = [ 'fairface' ]
+	model_source_set = get_model_options().get('sources')
+
+	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)
+
+
+def clear_inference_pool() -> None:
+	model_names = [ 'fairface' ]
+	inference_manager.clear_inference_pool(__name__, model_names)
+
+
+def get_model_options() -> ModelOptions:
+	return create_static_model_set('full').get('fairface')
+
+
+def pre_check() -> bool:
+	model_hash_set = get_model_options().get('hashes')
+	model_source_set = get_model_options().get('sources')
+
+	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)
+
+
+def classify_face(temp_vision_frame : VisionFrame, face_landmark_5 : FaceLandmark5) -> Tuple[Gender, Age, Race]:
+	model_template = get_model_options().get('template')
+	model_size = get_model_options().get('size')
+	model_mean = get_model_options().get('mean')
+	model_standard_deviation = get_model_options().get('standard_deviation')
+	crop_vision_frame, _ = warp_face_by_face_landmark_5(temp_vision_frame, face_landmark_5, model_template, model_size)
+	crop_vision_frame = crop_vision_frame.astype(numpy.float32)[:, :, ::-1] / 255.0
+	crop_vision_frame -= model_mean
+	crop_vision_frame /= model_standard_deviation
+	crop_vision_frame = crop_vision_frame.transpose(2, 0, 1)
+	crop_vision_frame = numpy.expand_dims(crop_vision_frame, axis = 0)
+	gender_id, age_id, race_id = forward(crop_vision_frame)
+	gender = categorize_gender(gender_id[0])
+	age = categorize_age(age_id[0])
+	race = categorize_race(race_id[0])
+	return gender, age, race
+
+
+def forward(crop_vision_frame : VisionFrame) -> Tuple[List[int], List[int], List[int]]:
+	face_classifier = get_inference_pool().get('face_classifier')
+
+	with conditional_thread_semaphore():
+		race_id, gender_id, age_id = face_classifier.run(None,
+		{
+			'input': crop_vision_frame
+		})
+
+	return gender_id, age_id, race_id
+
+
+def categorize_gender(gender_id : int) -> Gender:
+	if gender_id == 1:
+		return 'female'
+	return 'male'
+
+
+def categorize_age(age_id : int) -> Age:
+	if age_id == 0:
+		return range(0, 2)
+	if age_id == 1:
+		return range(3, 9)
+	if age_id == 2:
+		return range(10, 19)
+	if age_id == 3:
+		return range(20, 29)
+	if age_id == 4:
+		return range(30, 39)
+	if age_id == 5:
+		return range(40, 49)
+	if age_id == 6:
+		return range(50, 59)
+	if age_id == 7:
+		return range(60, 69)
+	return range(70, 100)
+
+
+def categorize_race(race_id : int) -> Race:
+	if race_id == 1:
+		return 'black'
+	if race_id == 2:
+		return 'latino'
+	if race_id == 3 or race_id == 4:
+		return 'asian'
+	if race_id == 5:
+		return 'indian'
+	if race_id == 6:
+		return 'arabic'
+	return 'white'

facefusion/face_detector.py

@@ -0,0 +1,323 @@
+from functools import lru_cache
+from typing import List, Sequence, Tuple
+
+import cv2
+import numpy
+
+from facefusion import inference_manager, state_manager
+from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
+from facefusion.face_helper import create_rotated_matrix_and_size, create_static_anchors, distance_to_bounding_box, distance_to_face_landmark_5, normalize_bounding_box, transform_bounding_box, transform_points
+from facefusion.filesystem import resolve_relative_path
+from facefusion.thread_helper import thread_semaphore
+from facefusion.types import Angle, BoundingBox, Detection, DownloadScope, DownloadSet, FaceLandmark5, InferencePool, ModelSet, Score, VisionFrame
+from facefusion.vision import restrict_frame, unpack_resolution
+
+
+@lru_cache(maxsize = None)
+def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
+	return\
+	{
+		'retinaface':
+		{
+			'hashes':
+			{
+				'retinaface':
+				{
+					'url': resolve_download_url('models-3.0.0', 'retinaface_10g.hash'),
+					'path': resolve_relative_path('../.assets/models/retinaface_10g.hash')
+				}
+			},
+			'sources':
+			{
+				'retinaface':
+				{
+					'url': resolve_download_url('models-3.0.0', 'retinaface_10g.onnx'),
+					'path': resolve_relative_path('../.assets/models/retinaface_10g.onnx')
+				}
+			}
+		},
+		'scrfd':
+		{
+			'hashes':
+			{
+				'scrfd':
+				{
+					'url': resolve_download_url('models-3.0.0', 'scrfd_2.5g.hash'),
+					'path': resolve_relative_path('../.assets/models/scrfd_2.5g.hash')
+				}
+			},
+			'sources':
+			{
+				'scrfd':
+				{
+					'url': resolve_download_url('models-3.0.0', 'scrfd_2.5g.onnx'),
+					'path': resolve_relative_path('../.assets/models/scrfd_2.5g.onnx')
+				}
+			}
+		},
+		'yolo_face':
+		{
+			'hashes':
+			{
+				'yolo_face':
+				{
+					'url': resolve_download_url('models-3.0.0', 'yoloface_8n.hash'),
+					'path': resolve_relative_path('../.assets/models/yoloface_8n.hash')
+				}
+			},
+			'sources':
+			{
+				'yolo_face':
+				{
+					'url': resolve_download_url('models-3.0.0', 'yoloface_8n.onnx'),
+					'path': resolve_relative_path('../.assets/models/yoloface_8n.onnx')
+				}
+			}
+		}
+	}
+
+
+def get_inference_pool() -> InferencePool:
+	model_names = [ state_manager.get_item('face_detector_model') ]
+	_, model_source_set = collect_model_downloads()
+
+	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)
+
+
+def clear_inference_pool() -> None:
+	model_names = [ state_manager.get_item('face_detector_model') ]
+	inference_manager.clear_inference_pool(__name__, model_names)
+
+
+def collect_model_downloads() -> Tuple[DownloadSet, DownloadSet]:
+	model_set = create_static_model_set('full')
+	model_hash_set = {}
+	model_source_set = {}
+
+	for face_detector_model in [ 'retinaface', 'scrfd', 'yolo_face' ]:
+		if state_manager.get_item('face_detector_model') in [ 'many', face_detector_model ]:
+			model_hash_set[face_detector_model] = model_set.get(face_detector_model).get('hashes').get(face_detector_model)
+			model_source_set[face_detector_model] = model_set.get(face_detector_model).get('sources').get(face_detector_model)
+
+	return model_hash_set, model_source_set
+
+
+def pre_check() -> bool:
+	model_hash_set, model_source_set = collect_model_downloads()
+
+	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)
+
+
+def detect_faces(vision_frame : VisionFrame) -> Tuple[List[BoundingBox], List[Score], List[FaceLandmark5]]:
+	all_bounding_boxes : List[BoundingBox] = []
+	all_face_scores : List[Score] = []
+	all_face_landmarks_5 : List[FaceLandmark5] = []
+
+	if state_manager.get_item('face_detector_model') in [ 'many', 'retinaface' ]:
+		bounding_boxes, face_scores, face_landmarks_5 = detect_with_retinaface(vision_frame, state_manager.get_item('face_detector_size'))
+		all_bounding_boxes.extend(bounding_boxes)
+		all_face_scores.extend(face_scores)
+		all_face_landmarks_5.extend(face_landmarks_5)
+
+	if state_manager.get_item('face_detector_model') in [ 'many', 'scrfd' ]:
+		bounding_boxes, face_scores, face_landmarks_5 = detect_with_scrfd(vision_frame, state_manager.get_item('face_detector_size'))
+		all_bounding_boxes.extend(bounding_boxes)
+		all_face_scores.extend(face_scores)
+		all_face_landmarks_5.extend(face_landmarks_5)
+
+	if state_manager.get_item('face_detector_model') in [ 'many', 'yolo_face' ]:
+		bounding_boxes, face_scores, face_landmarks_5 = detect_with_yolo_face(vision_frame, state_manager.get_item('face_detector_size'))
+		all_bounding_boxes.extend(bounding_boxes)
+		all_face_scores.extend(face_scores)
+		all_face_landmarks_5.extend(face_landmarks_5)
+
+	all_bounding_boxes = [ normalize_bounding_box(all_bounding_box) for all_bounding_box in all_bounding_boxes ]
+	return all_bounding_boxes, all_face_scores, all_face_landmarks_5
+
+
+def detect_rotated_faces(vision_frame : VisionFrame, angle : Angle) -> Tuple[List[BoundingBox], List[Score], List[FaceLandmark5]]:
+	rotated_matrix, rotated_size = create_rotated_matrix_and_size(angle, vision_frame.shape[:2][::-1])
+	rotated_vision_frame = cv2.warpAffine(vision_frame, rotated_matrix, rotated_size)
+	rotated_inverse_matrix = cv2.invertAffineTransform(rotated_matrix)
+	bounding_boxes, face_scores, face_landmarks_5 = detect_faces(rotated_vision_frame)
+	bounding_boxes = [ transform_bounding_box(bounding_box, rotated_inverse_matrix) for bounding_box in bounding_boxes ]
+	face_landmarks_5 = [ transform_points(face_landmark_5, rotated_inverse_matrix) for face_landmark_5 in face_landmarks_5 ]
+	return bounding_boxes, face_scores, face_landmarks_5
+
+
+def detect_with_retinaface(vision_frame : VisionFrame, face_detector_size : str) -> Tuple[List[BoundingBox], List[Score], List[FaceLandmark5]]:
+	bounding_boxes = []
+	face_scores = []
+	face_landmarks_5 = []
+	feature_strides = [ 8, 16, 32 ]
+	feature_map_channel = 3
+	anchor_total = 2
+	face_detector_score = state_manager.get_item('face_detector_score')
+	face_detector_width, face_detector_height = unpack_resolution(face_detector_size)
+	temp_vision_frame = restrict_frame(vision_frame, (face_detector_width, face_detector_height))
+	ratio_height = vision_frame.shape[0] / temp_vision_frame.shape[0]
+	ratio_width = vision_frame.shape[1] / temp_vision_frame.shape[1]
+	detect_vision_frame = prepare_detect_frame(temp_vision_frame, face_detector_size)
+	detect_vision_frame = normalize_detect_frame(detect_vision_frame, [ -1, 1 ])
+	detection = forward_with_retinaface(detect_vision_frame)
+
+	for index, feature_stride in enumerate(feature_strides):
+		keep_indices = numpy.where(detection[index] >= face_detector_score)[0]
+
+		if numpy.any(keep_indices):
+			stride_height = face_detector_height // feature_stride
+			stride_width = face_detector_width // feature_stride
+			anchors = create_static_anchors(feature_stride, anchor_total, stride_height, stride_width)
+			bounding_boxes_raw = detection[index + feature_map_channel] * feature_stride
+			face_landmarks_5_raw = detection[index + feature_map_channel * 2] * feature_stride
+
+			for bounding_box_raw in distance_to_bounding_box(anchors, bounding_boxes_raw)[keep_indices]:
+				bounding_boxes.append(numpy.array(
+				[
+					bounding_box_raw[0] * ratio_width,
+					bounding_box_raw[1] * ratio_height,
+					bounding_box_raw[2] * ratio_width,
+					bounding_box_raw[3] * ratio_height
+				]))
+
+			for face_score_raw in detection[index][keep_indices]:
+				face_scores.append(face_score_raw[0])
+
+			for face_landmark_raw_5 in distance_to_face_landmark_5(anchors, face_landmarks_5_raw)[keep_indices]:
+				face_landmarks_5.append(face_landmark_raw_5 * [ ratio_width, ratio_height ])
+
+	return bounding_boxes, face_scores, face_landmarks_5
+
+
+def detect_with_scrfd(vision_frame : VisionFrame, face_detector_size : str) -> Tuple[List[BoundingBox], List[Score], List[FaceLandmark5]]:
+	bounding_boxes = []
+	face_scores = []
+	face_landmarks_5 = []
+	feature_strides = [ 8, 16, 32 ]
+	feature_map_channel = 3
+	anchor_total = 2
+	face_detector_score = state_manager.get_item('face_detector_score')
+	face_detector_width, face_detector_height = unpack_resolution(face_detector_size)
+	temp_vision_frame = restrict_frame(vision_frame, (face_detector_width, face_detector_height))
+	ratio_height = vision_frame.shape[0] / temp_vision_frame.shape[0]
+	ratio_width = vision_frame.shape[1] / temp_vision_frame.shape[1]
+	detect_vision_frame = prepare_detect_frame(temp_vision_frame, face_detector_size)
+	detect_vision_frame = normalize_detect_frame(detect_vision_frame, [ -1, 1 ])
+	detection = forward_with_scrfd(detect_vision_frame)
+
+	for index, feature_stride in enumerate(feature_strides):
+		keep_indices = numpy.where(detection[index] >= face_detector_score)[0]
+
+		if numpy.any(keep_indices):
+			stride_height = face_detector_height // feature_stride
+			stride_width = face_detector_width // feature_stride
+			anchors = create_static_anchors(feature_stride, anchor_total, stride_height, stride_width)
+			bounding_boxes_raw = detection[index + feature_map_channel] * feature_stride
+			face_landmarks_5_raw = detection[index + feature_map_channel * 2] * feature_stride
+
+			for bounding_box_raw in distance_to_bounding_box(anchors, bounding_boxes_raw)[keep_indices]:
+				bounding_boxes.append(numpy.array(
+				[
+					bounding_box_raw[0] * ratio_width,
+					bounding_box_raw[1] * ratio_height,
+					bounding_box_raw[2] * ratio_width,
+					bounding_box_raw[3] * ratio_height
+				]))
+
+			for face_score_raw in detection[index][keep_indices]:
+				face_scores.append(face_score_raw[0])
+
+			for face_landmark_raw_5 in distance_to_face_landmark_5(anchors, face_landmarks_5_raw)[keep_indices]:
+				face_landmarks_5.append(face_landmark_raw_5 * [ ratio_width, ratio_height ])
+
+	return bounding_boxes, face_scores, face_landmarks_5
+
+
+def detect_with_yolo_face(vision_frame : VisionFrame, face_detector_size : str) -> Tuple[List[BoundingBox], List[Score], List[FaceLandmark5]]:
+	bounding_boxes = []
+	face_scores = []
+	face_landmarks_5 = []
+	face_detector_score = state_manager.get_item('face_detector_score')
+	face_detector_width, face_detector_height = unpack_resolution(face_detector_size)
+	temp_vision_frame = restrict_frame(vision_frame, (face_detector_width, face_detector_height))
+	ratio_height = vision_frame.shape[0] / temp_vision_frame.shape[0]
+	ratio_width = vision_frame.shape[1] / temp_vision_frame.shape[1]
+	detect_vision_frame = prepare_detect_frame(temp_vision_frame, face_detector_size)
+	detect_vision_frame = normalize_detect_frame(detect_vision_frame, [ 0, 1 ])
+	detection = forward_with_yolo_face(detect_vision_frame)
+	detection = numpy.squeeze(detection).T
+	bounding_boxes_raw, face_scores_raw, face_landmarks_5_raw = numpy.split(detection, [ 4, 5 ], axis = 1)
+	keep_indices = numpy.where(face_scores_raw > face_detector_score)[0]
+
+	if numpy.any(keep_indices):
+		bounding_boxes_raw, face_scores_raw, face_landmarks_5_raw = bounding_boxes_raw[keep_indices], face_scores_raw[keep_indices], face_landmarks_5_raw[keep_indices]
+
+		for bounding_box_raw in bounding_boxes_raw:
+			bounding_boxes.append(numpy.array(
+			[
+				(bounding_box_raw[0] - bounding_box_raw[2] / 2) * ratio_width,
+				(bounding_box_raw[1] - bounding_box_raw[3] / 2) * ratio_height,
+				(bounding_box_raw[0] + bounding_box_raw[2] / 2) * ratio_width,
+				(bounding_box_raw[1] + bounding_box_raw[3] / 2) * ratio_height
+			]))
+
+		face_scores = face_scores_raw.ravel().tolist()
+		face_landmarks_5_raw[:, 0::3] = (face_landmarks_5_raw[:, 0::3]) * ratio_width
+		face_landmarks_5_raw[:, 1::3] = (face_landmarks_5_raw[:, 1::3]) * ratio_height
+
+		for face_landmark_raw_5 in face_landmarks_5_raw:
+			face_landmarks_5.append(numpy.array(face_landmark_raw_5.reshape(-1, 3)[:, :2]))
+
+	return bounding_boxes, face_scores, face_landmarks_5
+
+
+def forward_with_retinaface(detect_vision_frame : VisionFrame) -> Detection:
+	face_detector = get_inference_pool().get('retinaface')
+
+	with thread_semaphore():
+		detection = face_detector.run(None,
+		{
+			'input': detect_vision_frame
+		})
+
+	return detection
+
+
+def forward_with_scrfd(detect_vision_frame : VisionFrame) -> Detection:
+	face_detector = get_inference_pool().get('scrfd')
+
+	with thread_semaphore():
+		detection = face_detector.run(None,
+		{
+			'input': detect_vision_frame
+		})
+
+	return detection
+
+
+def forward_with_yolo_face(detect_vision_frame : VisionFrame) -> Detection:
+	face_detector = get_inference_pool().get('yolo_face')
+
+	with thread_semaphore():
+		detection = face_detector.run(None,
+		{
+			'input': detect_vision_frame
+		})
+
+	return detection
+
+
+def prepare_detect_frame(temp_vision_frame : VisionFrame, face_detector_size : str) -> VisionFrame:
+	face_detector_width, face_detector_height = unpack_resolution(face_detector_size)
+	detect_vision_frame = numpy.zeros((face_detector_height, face_detector_width, 3))
+	detect_vision_frame[:temp_vision_frame.shape[0], :temp_vision_frame.shape[1], :] = temp_vision_frame
+	detect_vision_frame = numpy.expand_dims(detect_vision_frame.transpose(2, 0, 1), axis = 0).astype(numpy.float32)
+	return detect_vision_frame
+
+
+def normalize_detect_frame(detect_vision_frame : VisionFrame, normalize_range : Sequence[int]) -> VisionFrame:
+	if normalize_range == [ -1, 1 ]:
+		return (detect_vision_frame - 127.5) / 128.0
+	if normalize_range == [ 0, 1 ]:
+		return detect_vision_frame / 255.0
+	return detect_vision_frame

facefusion/face_helper.py

@@ -0,0 +1,254 @@
+from functools import lru_cache
+from typing import List, Sequence, Tuple
+
+import cv2
+import numpy
+from cv2.typing import Size
+
+from facefusion.types import Anchors, Angle, BoundingBox, Distance, FaceDetectorModel, FaceLandmark5, FaceLandmark68, Mask, Matrix, Points, Scale, Score, Translation, VisionFrame, WarpTemplate, WarpTemplateSet
+
+WARP_TEMPLATE_SET : WarpTemplateSet =\
+{
+	'arcface_112_v1': numpy.array(
+	[
+		[ 0.35473214, 0.45658929 ],
+		[ 0.64526786, 0.45658929 ],
+		[ 0.50000000, 0.61154464 ],
+		[ 0.37913393, 0.77687500 ],
+		[ 0.62086607, 0.77687500 ]
+	]),
+	'arcface_112_v2': numpy.array(
+	[
+		[ 0.34191607, 0.46157411 ],
+		[ 0.65653393, 0.45983393 ],
+		[ 0.50022500, 0.64050536 ],
+		[ 0.37097589, 0.82469196 ],
+		[ 0.63151696, 0.82325089 ]
+	]),
+	'arcface_128': numpy.array(
+	[
+		[ 0.36167656, 0.40387734 ],
+		[ 0.63696719, 0.40235469 ],
+		[ 0.50019687, 0.56044219 ],
+		[ 0.38710391, 0.72160547 ],
+		[ 0.61507734, 0.72034453 ]
+	]),
+	'dfl_whole_face': numpy.array(
+	[
+		[ 0.35342266, 0.39285716 ],
+		[ 0.62797622, 0.39285716 ],
+		[ 0.48660713, 0.54017860 ],
+		[ 0.38839287, 0.68750011 ],
+		[ 0.59821427, 0.68750011 ]
+	]),
+	'ffhq_512': numpy.array(
+	[
+		[ 0.37691676, 0.46864664 ],
+		[ 0.62285697, 0.46912813 ],
+		[ 0.50123859, 0.61331904 ],
+		[ 0.39308822, 0.72541100 ],
+		[ 0.61150205, 0.72490465 ]
+	]),
+	'mtcnn_512': numpy.array(
+	[
+		[ 0.36562865, 0.46733799 ],
+		[ 0.63305391, 0.46585885 ],
+		[ 0.50019127, 0.61942959 ],
+		[ 0.39032951, 0.77598822 ],
+		[ 0.61178945, 0.77476328 ]
+	]),
+	'styleganex_384': numpy.array(
+	[
+		[ 0.42353745, 0.52289879 ],
+		[ 0.57725008, 0.52319972 ],
+		[ 0.50123859, 0.61331904 ],
+		[ 0.43364461, 0.68337652 ],
+		[ 0.57015325, 0.68306005 ]
+	])
+}
+
+
+def estimate_matrix_by_face_landmark_5(face_landmark_5 : FaceLandmark5, warp_template : WarpTemplate, crop_size : Size) -> Matrix:
+	normed_warp_template = WARP_TEMPLATE_SET.get(warp_template) * crop_size
+	affine_matrix = cv2.estimateAffinePartial2D(face_landmark_5, normed_warp_template, method = cv2.RANSAC, ransacReprojThreshold = 100)[0]
+	return affine_matrix
+
+
+def warp_face_by_face_landmark_5(temp_vision_frame : VisionFrame, face_landmark_5 : FaceLandmark5, warp_template : WarpTemplate, crop_size : Size) -> Tuple[VisionFrame, Matrix]:
+	affine_matrix = estimate_matrix_by_face_landmark_5(face_landmark_5, warp_template, crop_size)
+	crop_vision_frame = cv2.warpAffine(temp_vision_frame, affine_matrix, crop_size, borderMode = cv2.BORDER_REPLICATE, flags = cv2.INTER_AREA)
+	return crop_vision_frame, affine_matrix
+
+
+def warp_face_by_bounding_box(temp_vision_frame : VisionFrame, bounding_box : BoundingBox, crop_size : Size) -> Tuple[VisionFrame, Matrix]:
+	source_points = numpy.array([ [ bounding_box[0], bounding_box[1] ], [bounding_box[2], bounding_box[1] ], [ bounding_box[0], bounding_box[3] ] ]).astype(numpy.float32)
+	target_points = numpy.array([ [ 0, 0 ], [ crop_size[0], 0 ], [ 0, crop_size[1] ] ]).astype(numpy.float32)
+	affine_matrix = cv2.getAffineTransform(source_points, target_points)
+	if bounding_box[2] - bounding_box[0] > crop_size[0] or bounding_box[3] - bounding_box[1] > crop_size[1]:
+		interpolation_method = cv2.INTER_AREA
+	else:
+		interpolation_method = cv2.INTER_LINEAR
+	crop_vision_frame = cv2.warpAffine(temp_vision_frame, affine_matrix, crop_size, flags = interpolation_method)
+	return crop_vision_frame, affine_matrix
+
+
+def warp_face_by_translation(temp_vision_frame : VisionFrame, translation : Translation, scale : float, crop_size : Size) -> Tuple[VisionFrame, Matrix]:
+	affine_matrix = numpy.array([ [ scale, 0, translation[0] ], [ 0, scale, translation[1] ] ])
+	crop_vision_frame = cv2.warpAffine(temp_vision_frame, affine_matrix, crop_size)
+	return crop_vision_frame, affine_matrix
+
+
+def paste_back(temp_vision_frame : VisionFrame, crop_vision_frame : VisionFrame, crop_mask : Mask, affine_matrix : Matrix) -> VisionFrame:
+	paste_bounding_box, paste_matrix = calc_paste_area(temp_vision_frame, crop_vision_frame, affine_matrix)
+	x_min, y_min, x_max, y_max = paste_bounding_box
+	paste_width = x_max - x_min
+	paste_height = y_max - y_min
+	inverse_mask = cv2.warpAffine(crop_mask, paste_matrix, (paste_width, paste_height)).clip(0, 1)
+	inverse_mask = numpy.expand_dims(inverse_mask, axis = -1)
+	inverse_vision_frame = cv2.warpAffine(crop_vision_frame, paste_matrix, (paste_width, paste_height), borderMode = cv2.BORDER_REPLICATE)
+	temp_vision_frame = temp_vision_frame.copy()
+	paste_vision_frame = temp_vision_frame[y_min:y_max, x_min:x_max]
+	paste_vision_frame = paste_vision_frame * (1 - inverse_mask) + inverse_vision_frame * inverse_mask
+	temp_vision_frame[y_min:y_max, x_min:x_max] = paste_vision_frame.astype(temp_vision_frame.dtype)
+	return temp_vision_frame
+
+
+def calc_paste_area(temp_vision_frame : VisionFrame, crop_vision_frame : VisionFrame, affine_matrix : Matrix) -> Tuple[BoundingBox, Matrix]:
+	temp_height, temp_width = temp_vision_frame.shape[:2]
+	crop_height, crop_width = crop_vision_frame.shape[:2]
+	inverse_matrix = cv2.invertAffineTransform(affine_matrix)
+	crop_points = numpy.array([ [ 0, 0 ], [ crop_width, 0 ], [ crop_width, crop_height ], [ 0, crop_height ] ])
+	paste_region_points = transform_points(crop_points, inverse_matrix)
+	min_point = numpy.floor(paste_region_points.min(axis = 0)).astype(int)
+	max_point = numpy.ceil(paste_region_points.max(axis = 0)).astype(int)
+	x_min, y_min = numpy.clip(min_point, 0, [ temp_width, temp_height ])
+	x_max, y_max = numpy.clip(max_point, 0, [ temp_width, temp_height ])
+	paste_bounding_box = numpy.array([ x_min, y_min, x_max, y_max ])
+	paste_matrix = inverse_matrix.copy()
+	paste_matrix[0, 2] -= x_min
+	paste_matrix[1, 2] -= y_min
+	return paste_bounding_box, paste_matrix
+
+
+@lru_cache(maxsize = None)
+def create_static_anchors(feature_stride : int, anchor_total : int, stride_height : int, stride_width : int) -> Anchors:
+	y, x = numpy.mgrid[:stride_height, :stride_width][::-1]
+	anchors = numpy.stack((y, x), axis = -1)
+	anchors = (anchors * feature_stride).reshape((-1, 2))
+	anchors = numpy.stack([ anchors ] * anchor_total, axis = 1).reshape((-1, 2))
+	return anchors
+
+
+def create_rotated_matrix_and_size(angle : Angle, size : Size) -> Tuple[Matrix, Size]:
+	rotated_matrix = cv2.getRotationMatrix2D((size[0] / 2, size[1] / 2), angle, 1)
+	rotated_size = numpy.dot(numpy.abs(rotated_matrix[:, :2]), size)
+	rotated_matrix[:, -1] += (rotated_size - size) * 0.5 #type:ignore[misc]
+	rotated_size = int(rotated_size[0]), int(rotated_size[1])
+	return rotated_matrix, rotated_size
+
+
+def create_bounding_box(face_landmark_68 : FaceLandmark68) -> BoundingBox:
+	min_x, min_y = numpy.min(face_landmark_68, axis = 0)
+	max_x, max_y = numpy.max(face_landmark_68, axis = 0)
+	bounding_box = normalize_bounding_box(numpy.array([ min_x, min_y, max_x, max_y ]))
+	return bounding_box
+
+
+def normalize_bounding_box(bounding_box : BoundingBox) -> BoundingBox:
+	x1, y1, x2, y2 = bounding_box
+	x1, x2 = sorted([ x1, x2 ])
+	y1, y2 = sorted([ y1, y2 ])
+	return numpy.array([ x1, y1, x2, y2 ])
+
+
+def transform_points(points : Points, matrix : Matrix) -> Points:
+	points = points.reshape(-1, 1, 2)
+	points = cv2.transform(points, matrix) #type:ignore[assignment]
+	points = points.reshape(-1, 2)
+	return points
+
+
+def transform_bounding_box(bounding_box : BoundingBox, matrix : Matrix) -> BoundingBox:
+	points = numpy.array(
+	[
+		[ bounding_box[0], bounding_box[1] ],
+		[ bounding_box[2], bounding_box[1] ],
+		[ bounding_box[2], bounding_box[3] ],
+		[ bounding_box[0], bounding_box[3] ]
+	])
+	points = transform_points(points, matrix)
+	x1, y1 = numpy.min(points, axis = 0)
+	x2, y2 = numpy.max(points, axis = 0)
+	return normalize_bounding_box(numpy.array([ x1, y1, x2, y2 ]))
+
+
+def distance_to_bounding_box(points : Points, distance : Distance) -> BoundingBox:
+	x1 = points[:, 0] - distance[:, 0]
+	y1 = points[:, 1] - distance[:, 1]
+	x2 = points[:, 0] + distance[:, 2]
+	y2 = points[:, 1] + distance[:, 3]
+	bounding_box = numpy.column_stack([ x1, y1, x2, y2 ])
+	return bounding_box
+
+
+def distance_to_face_landmark_5(points : Points, distance : Distance) -> FaceLandmark5:
+	x = points[:, 0::2] + distance[:, 0::2]
+	y = points[:, 1::2] + distance[:, 1::2]
+	face_landmark_5 = numpy.stack((x, y), axis = -1)
+	return face_landmark_5
+
+
+def scale_face_landmark_5(face_landmark_5 : FaceLandmark5, scale : Scale) -> FaceLandmark5:
+	face_landmark_5_scale = face_landmark_5 - face_landmark_5[2]
+	face_landmark_5_scale *= scale
+	face_landmark_5_scale += face_landmark_5[2]
+	return face_landmark_5_scale
+
+
+def convert_to_face_landmark_5(face_landmark_68 : FaceLandmark68) -> FaceLandmark5:
+	face_landmark_5 = numpy.array(
+	[
+		numpy.mean(face_landmark_68[36:42], axis = 0),
+		numpy.mean(face_landmark_68[42:48], axis = 0),
+		face_landmark_68[30],
+		face_landmark_68[48],
+		face_landmark_68[54]
+	])
+	return face_landmark_5
+
+
+def estimate_face_angle(face_landmark_68 : FaceLandmark68) -> Angle:
+	x1, y1 = face_landmark_68[0]
+	x2, y2 = face_landmark_68[16]
+	theta = numpy.arctan2(y2 - y1, x2 - x1)
+	theta = numpy.degrees(theta) % 360
+	angles = numpy.linspace(0, 360, 5)
+	index = numpy.argmin(numpy.abs(angles - theta))
+	face_angle = int(angles[index] % 360)
+	return face_angle
+
+
+def apply_nms(bounding_boxes : List[BoundingBox], scores : List[Score], score_threshold : float, nms_threshold : float) -> Sequence[int]:
+	normed_bounding_boxes = [ (x1, y1, x2 - x1, y2 - y1) for (x1, y1, x2, y2) in bounding_boxes ]
+	keep_indices = cv2.dnn.NMSBoxes(normed_bounding_boxes, scores, score_threshold = score_threshold, nms_threshold = nms_threshold)
+	return keep_indices
+
+
+def get_nms_threshold(face_detector_model : FaceDetectorModel, face_detector_angles : List[Angle]) -> float:
+	if face_detector_model == 'many':
+		return 0.1
+	if len(face_detector_angles) == 2:
+		return 0.3
+	if len(face_detector_angles) == 3:
+		return 0.2
+	if len(face_detector_angles) == 4:
+		return 0.1
+	return 0.4
+
+
+def merge_matrix(matrices : List[Matrix]) -> Matrix:
+	merged_matrix = numpy.vstack([ matrices[0], [ 0, 0, 1 ] ])
+	for matrix in matrices[1:]:
+		matrix = numpy.vstack([ matrix, [ 0, 0, 1 ] ])
+		merged_matrix = numpy.dot(merged_matrix, matrix)
+	return merged_matrix[:2, :]

facefusion/face_landmarker.py

@@ -0,0 +1,222 @@
+from functools import lru_cache
+from typing import Tuple
+
+import cv2
+import numpy
+
+from facefusion import inference_manager, state_manager
+from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
+from facefusion.face_helper import create_rotated_matrix_and_size, estimate_matrix_by_face_landmark_5, transform_points, warp_face_by_translation
+from facefusion.filesystem import resolve_relative_path
+from facefusion.thread_helper import conditional_thread_semaphore
+from facefusion.types import Angle, BoundingBox, DownloadScope, DownloadSet, FaceLandmark5, FaceLandmark68, InferencePool, ModelSet, Prediction, Score, VisionFrame
+
+
+@lru_cache(maxsize = None)
+def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
+	return\
+	{
+		'2dfan4':
+		{
+			'hashes':
+			{
+				'2dfan4':
+				{
+					'url': resolve_download_url('models-3.0.0', '2dfan4.hash'),
+					'path': resolve_relative_path('../.assets/models/2dfan4.hash')
+				}
+			},
+			'sources':
+			{
+				'2dfan4':
+				{
+					'url': resolve_download_url('models-3.0.0', '2dfan4.onnx'),
+					'path': resolve_relative_path('../.assets/models/2dfan4.onnx')
+				}
+			},
+			'size': (256, 256)
+		},
+		'peppa_wutz':
+		{
+			'hashes':
+			{
+				'peppa_wutz':
+				{
+					'url': resolve_download_url('models-3.0.0', 'peppa_wutz.hash'),
+					'path': resolve_relative_path('../.assets/models/peppa_wutz.hash')
+				}
+			},
+			'sources':
+			{
+				'peppa_wutz':
+				{
+					'url': resolve_download_url('models-3.0.0', 'peppa_wutz.onnx'),
+					'path': resolve_relative_path('../.assets/models/peppa_wutz.onnx')
+				}
+			},
+			'size': (256, 256)
+		},
+		'fan_68_5':
+		{
+			'hashes':
+			{
+				'fan_68_5':
+				{
+					'url': resolve_download_url('models-3.0.0', 'fan_68_5.hash'),
+					'path': resolve_relative_path('../.assets/models/fan_68_5.hash')
+				}
+			},
+			'sources':
+			{
+				'fan_68_5':
+				{
+					'url': resolve_download_url('models-3.0.0', 'fan_68_5.onnx'),
+					'path': resolve_relative_path('../.assets/models/fan_68_5.onnx')
+				}
+			}
+		}
+	}
+
+
+def get_inference_pool() -> InferencePool:
+	model_names = [ state_manager.get_item('face_landmarker_model'), 'fan_68_5' ]
+	_, model_source_set = collect_model_downloads()
+
+	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)
+
+
+def clear_inference_pool() -> None:
+	model_names = [ state_manager.get_item('face_landmarker_model'), 'fan_68_5' ]
+	inference_manager.clear_inference_pool(__name__, model_names)
+
+
+def collect_model_downloads() -> Tuple[DownloadSet, DownloadSet]:
+	model_set = create_static_model_set('full')
+	model_hash_set =\
+	{
+		'fan_68_5': model_set.get('fan_68_5').get('hashes').get('fan_68_5')
+	}
+	model_source_set =\
+	{
+		'fan_68_5': model_set.get('fan_68_5').get('sources').get('fan_68_5')
+	}
+
+	for face_landmarker_model in [ '2dfan4', 'peppa_wutz' ]:
+		if state_manager.get_item('face_landmarker_model') in [ 'many', face_landmarker_model ]:
+			model_hash_set[face_landmarker_model] = model_set.get(face_landmarker_model).get('hashes').get(face_landmarker_model)
+			model_source_set[face_landmarker_model] = model_set.get(face_landmarker_model).get('sources').get(face_landmarker_model)
+
+	return model_hash_set, model_source_set
+
+
+def pre_check() -> bool:
+	model_hash_set, model_source_set = collect_model_downloads()
+
+	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)
+
+
+def detect_face_landmark(vision_frame : VisionFrame, bounding_box : BoundingBox, face_angle : Angle) -> Tuple[FaceLandmark68, Score]:
+	face_landmark_2dfan4 = None
+	face_landmark_peppa_wutz = None
+	face_landmark_score_2dfan4 = 0.0
+	face_landmark_score_peppa_wutz = 0.0
+
+	if state_manager.get_item('face_landmarker_model') in [ 'many', '2dfan4' ]:
+		face_landmark_2dfan4, face_landmark_score_2dfan4 = detect_with_2dfan4(vision_frame, bounding_box, face_angle)
+
+	if state_manager.get_item('face_landmarker_model') in [ 'many', 'peppa_wutz' ]:
+		face_landmark_peppa_wutz, face_landmark_score_peppa_wutz = detect_with_peppa_wutz(vision_frame, bounding_box, face_angle)
+
+	if face_landmark_score_2dfan4 > face_landmark_score_peppa_wutz - 0.2:
+		return face_landmark_2dfan4, face_landmark_score_2dfan4
+	return face_landmark_peppa_wutz, face_landmark_score_peppa_wutz
+
+
+def detect_with_2dfan4(temp_vision_frame: VisionFrame, bounding_box: BoundingBox, face_angle: Angle) -> Tuple[FaceLandmark68, Score]:
+	model_size = create_static_model_set('full').get('2dfan4').get('size')
+	scale = 195 / numpy.subtract(bounding_box[2:], bounding_box[:2]).max().clip(1, None)
+	translation = (model_size[0] - numpy.add(bounding_box[2:], bounding_box[:2]) * scale) * 0.5
+	rotated_matrix, rotated_size = create_rotated_matrix_and_size(face_angle, model_size)
+	crop_vision_frame, affine_matrix = warp_face_by_translation(temp_vision_frame, translation, scale, model_size)
+	crop_vision_frame = cv2.warpAffine(crop_vision_frame, rotated_matrix, rotated_size)
+	crop_vision_frame = conditional_optimize_contrast(crop_vision_frame)
+	crop_vision_frame = crop_vision_frame.transpose(2, 0, 1).astype(numpy.float32) / 255.0
+	face_landmark_68, face_heatmap = forward_with_2dfan4(crop_vision_frame)
+	face_landmark_68 = face_landmark_68[:, :, :2][0] / 64 * 256
+	face_landmark_68 = transform_points(face_landmark_68, cv2.invertAffineTransform(rotated_matrix))
+	face_landmark_68 = transform_points(face_landmark_68, cv2.invertAffineTransform(affine_matrix))
+	face_landmark_score_68 = numpy.amax(face_heatmap, axis = (2, 3))
+	face_landmark_score_68 = numpy.mean(face_landmark_score_68)
+	face_landmark_score_68 = numpy.interp(face_landmark_score_68, [ 0, 0.9 ], [ 0, 1 ])
+	return face_landmark_68, face_landmark_score_68
+
+
+def detect_with_peppa_wutz(temp_vision_frame : VisionFrame, bounding_box : BoundingBox, face_angle : Angle) -> Tuple[FaceLandmark68, Score]:
+	model_size = create_static_model_set('full').get('peppa_wutz').get('size')
+	scale = 195 / numpy.subtract(bounding_box[2:], bounding_box[:2]).max().clip(1, None)
+	translation = (model_size[0] - numpy.add(bounding_box[2:], bounding_box[:2]) * scale) * 0.5
+	rotated_matrix, rotated_size = create_rotated_matrix_and_size(face_angle, model_size)
+	crop_vision_frame, affine_matrix = warp_face_by_translation(temp_vision_frame, translation, scale, model_size)
+	crop_vision_frame = cv2.warpAffine(crop_vision_frame, rotated_matrix, rotated_size)
+	crop_vision_frame = conditional_optimize_contrast(crop_vision_frame)
+	crop_vision_frame = crop_vision_frame.transpose(2, 0, 1).astype(numpy.float32) / 255.0
+	crop_vision_frame = numpy.expand_dims(crop_vision_frame, axis = 0)
+	prediction = forward_with_peppa_wutz(crop_vision_frame)
+	face_landmark_68 = prediction.reshape(-1, 3)[:, :2] / 64 * model_size[0]
+	face_landmark_68 = transform_points(face_landmark_68, cv2.invertAffineTransform(rotated_matrix))
+	face_landmark_68 = transform_points(face_landmark_68, cv2.invertAffineTransform(affine_matrix))
+	face_landmark_score_68 = prediction.reshape(-1, 3)[:, 2].mean()
+	face_landmark_score_68 = numpy.interp(face_landmark_score_68, [ 0, 0.95 ], [ 0, 1 ])
+	return face_landmark_68, face_landmark_score_68
+
+
+def conditional_optimize_contrast(crop_vision_frame : VisionFrame) -> VisionFrame:
+	crop_vision_frame = cv2.cvtColor(crop_vision_frame, cv2.COLOR_RGB2Lab)
+	if numpy.mean(crop_vision_frame[:, :, 0]) < 30: #type:ignore[arg-type]
+		crop_vision_frame[:, :, 0] = cv2.createCLAHE(clipLimit = 2).apply(crop_vision_frame[:, :, 0])
+	crop_vision_frame = cv2.cvtColor(crop_vision_frame, cv2.COLOR_Lab2RGB)
+	return crop_vision_frame
+
+
+def estimate_face_landmark_68_5(face_landmark_5 : FaceLandmark5) -> FaceLandmark68:
+	affine_matrix = estimate_matrix_by_face_landmark_5(face_landmark_5, 'ffhq_512', (1, 1))
+	face_landmark_5 = cv2.transform(face_landmark_5.reshape(1, -1, 2), affine_matrix).reshape(-1, 2)
+	face_landmark_68_5 = forward_fan_68_5(face_landmark_5)
+	face_landmark_68_5 = cv2.transform(face_landmark_68_5.reshape(1, -1, 2), cv2.invertAffineTransform(affine_matrix)).reshape(-1, 2)
+	return face_landmark_68_5
+
+
+def forward_with_2dfan4(crop_vision_frame : VisionFrame) -> Tuple[Prediction, Prediction]:
+	face_landmarker = get_inference_pool().get('2dfan4')
+
+	with conditional_thread_semaphore():
+		prediction = face_landmarker.run(None,
+		{
+			'input': [ crop_vision_frame ]
+		})
+
+	return prediction
+
+
+def forward_with_peppa_wutz(crop_vision_frame : VisionFrame) -> Prediction:
+	face_landmarker = get_inference_pool().get('peppa_wutz')
+
+	with conditional_thread_semaphore():
+		prediction = face_landmarker.run(None,
+		{
+			'input': crop_vision_frame
+		})[0]
+
+	return prediction
+
+
+def forward_fan_68_5(face_landmark_5 : FaceLandmark5) -> FaceLandmark68:
+	face_landmarker = get_inference_pool().get('fan_68_5')
+
+	with conditional_thread_semaphore():
+		face_landmark_68_5 = face_landmarker.run(None,
+		{
+			'input': [ face_landmark_5 ]
+		})[0][0]
+
+	return face_landmark_68_5

facefusion/face_masker.py

@@ -0,0 +1,240 @@
+from functools import lru_cache
+from typing import List, Tuple
+
+import cv2
+import numpy
+
+import facefusion.choices
+from facefusion import inference_manager, state_manager
+from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
+from facefusion.filesystem import resolve_relative_path
+from facefusion.thread_helper import conditional_thread_semaphore
+from facefusion.types import DownloadScope, DownloadSet, FaceLandmark68, FaceMaskArea, FaceMaskRegion, InferencePool, Mask, ModelSet, Padding, VisionFrame
+
+
+@lru_cache(maxsize = None)
+def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
+	return\
+	{
+		'xseg_1':
+		{
+			'hashes':
+			{
+				'face_occluder':
+				{
+					'url': resolve_download_url('models-3.1.0', 'xseg_1.hash'),
+					'path': resolve_relative_path('../.assets/models/xseg_1.hash')
+				}
+			},
+			'sources':
+			{
+				'face_occluder':
+				{
+					'url': resolve_download_url('models-3.1.0', 'xseg_1.onnx'),
+					'path': resolve_relative_path('../.assets/models/xseg_1.onnx')
+				}
+			},
+			'size': (256, 256)
+		},
+		'xseg_2':
+		{
+			'hashes':
+			{
+				'face_occluder':
+				{
+					'url': resolve_download_url('models-3.1.0', 'xseg_2.hash'),
+					'path': resolve_relative_path('../.assets/models/xseg_2.hash')
+				}
+			},
+			'sources':
+			{
+				'face_occluder':
+				{
+					'url': resolve_download_url('models-3.1.0', 'xseg_2.onnx'),
+					'path': resolve_relative_path('../.assets/models/xseg_2.onnx')
+				}
+			},
+			'size': (256, 256)
+		},
+		'xseg_3':
+		{
+			'hashes':
+			{
+				'face_occluder':
+				{
+					'url': resolve_download_url('models-3.2.0', 'xseg_3.hash'),
+					'path': resolve_relative_path('../.assets/models/xseg_3.hash')
+				}
+			},
+			'sources':
+			{
+				'face_occluder':
+				{
+					'url': resolve_download_url('models-3.2.0', 'xseg_3.onnx'),
+					'path': resolve_relative_path('../.assets/models/xseg_3.onnx')
+				}
+			},
+			'size': (256, 256)
+		},
+		'bisenet_resnet_18':
+		{
+			'hashes':
+			{
+				'face_parser':
+				{
+					'url': resolve_download_url('models-3.1.0', 'bisenet_resnet_18.hash'),
+					'path': resolve_relative_path('../.assets/models/bisenet_resnet_18.hash')
+				}
+			},
+			'sources':
+			{
+				'face_parser':
+				{
+					'url': resolve_download_url('models-3.1.0', 'bisenet_resnet_18.onnx'),
+					'path': resolve_relative_path('../.assets/models/bisenet_resnet_18.onnx')
+				}
+			},
+			'size': (512, 512)
+		},
+		'bisenet_resnet_34':
+		{
+			'hashes':
+			{
+				'face_parser':
+				{
+					'url': resolve_download_url('models-3.0.0', 'bisenet_resnet_34.hash'),
+					'path': resolve_relative_path('../.assets/models/bisenet_resnet_34.hash')
+				}
+			},
+			'sources':
+			{
+				'face_parser':
+				{
+					'url': resolve_download_url('models-3.0.0', 'bisenet_resnet_34.onnx'),
+					'path': resolve_relative_path('../.assets/models/bisenet_resnet_34.onnx')
+				}
+			},
+			'size': (512, 512)
+		}
+	}
+
+
+def get_inference_pool() -> InferencePool:
+	model_names = [ state_manager.get_item('face_occluder_model'), state_manager.get_item('face_parser_model') ]
+	_, model_source_set = collect_model_downloads()
+
+	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)
+
+
+def clear_inference_pool() -> None:
+	model_names = [ state_manager.get_item('face_occluder_model'), state_manager.get_item('face_parser_model') ]
+	inference_manager.clear_inference_pool(__name__, model_names)
+
+
+def collect_model_downloads() -> Tuple[DownloadSet, DownloadSet]:
+	model_set = create_static_model_set('full')
+	model_hash_set = {}
+	model_source_set = {}
+
+	for face_occluder_model in [ 'xseg_1', 'xseg_2', 'xseg_3' ]:
+		if state_manager.get_item('face_occluder_model') == face_occluder_model:
+			model_hash_set[face_occluder_model] = model_set.get(face_occluder_model).get('hashes').get('face_occluder')
+			model_source_set[face_occluder_model] = model_set.get(face_occluder_model).get('sources').get('face_occluder')
+
+	for face_parser_model in [ 'bisenet_resnet_18', 'bisenet_resnet_34' ]:
+		if state_manager.get_item('face_parser_model') == face_parser_model:
+			model_hash_set[face_parser_model] = model_set.get(face_parser_model).get('hashes').get('face_parser')
+			model_source_set[face_parser_model] = model_set.get(face_parser_model).get('sources').get('face_parser')
+
+	return model_hash_set, model_source_set
+
+
+def pre_check() -> bool:
+	model_hash_set, model_source_set = collect_model_downloads()
+
+	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)
+
+
+def create_box_mask(crop_vision_frame : VisionFrame, face_mask_blur : float, face_mask_padding : Padding) -> Mask:
+	crop_size = crop_vision_frame.shape[:2][::-1]
+	blur_amount = int(crop_size[0] * 0.5 * face_mask_blur)
+	blur_area = max(blur_amount // 2, 1)
+	box_mask : Mask = numpy.ones(crop_size).astype(numpy.float32)
+	box_mask[:max(blur_area, int(crop_size[1] * face_mask_padding[0] / 100)), :] = 0
+	box_mask[-max(blur_area, int(crop_size[1] * face_mask_padding[2] / 100)):, :] = 0
+	box_mask[:, :max(blur_area, int(crop_size[0] * face_mask_padding[3] / 100))] = 0
+	box_mask[:, -max(blur_area, int(crop_size[0] * face_mask_padding[1] / 100)):] = 0
+
+	if blur_amount > 0:
+		box_mask = cv2.GaussianBlur(box_mask, (0, 0), blur_amount * 0.25)
+	return box_mask
+
+
+def create_occlusion_mask(crop_vision_frame : VisionFrame) -> Mask:
+	model_name = state_manager.get_item('face_occluder_model')
+	model_size = create_static_model_set('full').get(model_name).get('size')
+	prepare_vision_frame = cv2.resize(crop_vision_frame, model_size)
+	prepare_vision_frame = numpy.expand_dims(prepare_vision_frame, axis = 0).astype(numpy.float32) / 255.0
+	prepare_vision_frame = prepare_vision_frame.transpose(0, 1, 2, 3)
+	occlusion_mask = forward_occlude_face(prepare_vision_frame)
+	occlusion_mask = occlusion_mask.transpose(0, 1, 2).clip(0, 1).astype(numpy.float32)
+	occlusion_mask = cv2.resize(occlusion_mask, crop_vision_frame.shape[:2][::-1])
+	occlusion_mask = (cv2.GaussianBlur(occlusion_mask.clip(0, 1), (0, 0), 5).clip(0.5, 1) - 0.5) * 2
+	return occlusion_mask
+
+
+def create_area_mask(crop_vision_frame : VisionFrame, face_landmark_68 : FaceLandmark68, face_mask_areas : List[FaceMaskArea]) -> Mask:
+	crop_size = crop_vision_frame.shape[:2][::-1]
+	landmark_points = []
+
+	for face_mask_area in face_mask_areas:
+		if face_mask_area in facefusion.choices.face_mask_area_set:
+			landmark_points.extend(facefusion.choices.face_mask_area_set.get(face_mask_area))
+
+	convex_hull = cv2.convexHull(face_landmark_68[landmark_points].astype(numpy.int32))
+	area_mask = numpy.zeros(crop_size).astype(numpy.float32)
+	cv2.fillConvexPoly(area_mask, convex_hull, 1.0) # type: ignore[call-overload]
+	area_mask = (cv2.GaussianBlur(area_mask.clip(0, 1), (0, 0), 5).clip(0.5, 1) - 0.5) * 2
+	return area_mask
+
+
+def create_region_mask(crop_vision_frame : VisionFrame, face_mask_regions : List[FaceMaskRegion]) -> Mask:
+	model_name = state_manager.get_item('face_parser_model')
+	model_size = create_static_model_set('full').get(model_name).get('size')
+	prepare_vision_frame = cv2.resize(crop_vision_frame, model_size)
+	prepare_vision_frame = prepare_vision_frame[:, :, ::-1].astype(numpy.float32) / 255.0
+	prepare_vision_frame = numpy.subtract(prepare_vision_frame, numpy.array([ 0.485, 0.456, 0.406 ]).astype(numpy.float32))
+	prepare_vision_frame = numpy.divide(prepare_vision_frame, numpy.array([ 0.229, 0.224, 0.225 ]).astype(numpy.float32))
+	prepare_vision_frame = numpy.expand_dims(prepare_vision_frame, axis = 0)
+	prepare_vision_frame = prepare_vision_frame.transpose(0, 3, 1, 2)
+	region_mask = forward_parse_face(prepare_vision_frame)
+	region_mask = numpy.isin(region_mask.argmax(0), [ facefusion.choices.face_mask_region_set.get(face_mask_region) for face_mask_region in face_mask_regions ])
+	region_mask = cv2.resize(region_mask.astype(numpy.float32), crop_vision_frame.shape[:2][::-1])
+	region_mask = (cv2.GaussianBlur(region_mask.clip(0, 1), (0, 0), 5).clip(0.5, 1) - 0.5) * 2
+	return region_mask
+
+
+def forward_occlude_face(prepare_vision_frame : VisionFrame) -> Mask:
+	model_name = state_manager.get_item('face_occluder_model')
+	face_occluder = get_inference_pool().get(model_name)
+
+	with conditional_thread_semaphore():
+		occlusion_mask : Mask = face_occluder.run(None,
+		{
+			'input': prepare_vision_frame
+		})[0][0]
+
+	return occlusion_mask
+
+
+def forward_parse_face(prepare_vision_frame : VisionFrame) -> Mask:
+	model_name = state_manager.get_item('face_parser_model')
+	face_parser = get_inference_pool().get(model_name)
+
+	with conditional_thread_semaphore():
+		region_mask : Mask = face_parser.run(None,
+		{
+			'input': prepare_vision_frame
+		})[0][0]
+
+	return region_mask

facefusion/face_recognizer.py

@@ -0,0 +1,87 @@
+from functools import lru_cache
+from typing import Tuple
+
+import numpy
+
+from facefusion import inference_manager
+from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
+from facefusion.face_helper import warp_face_by_face_landmark_5
+from facefusion.filesystem import resolve_relative_path
+from facefusion.thread_helper import conditional_thread_semaphore
+from facefusion.types import DownloadScope, Embedding, FaceLandmark5, InferencePool, ModelOptions, ModelSet, VisionFrame
+
+
+@lru_cache(maxsize = None)
+def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
+	return\
+	{
+		'arcface':
+		{
+			'hashes':
+			{
+				'face_recognizer':
+				{
+					'url': resolve_download_url('models-3.0.0', 'arcface_w600k_r50.hash'),
+					'path': resolve_relative_path('../.assets/models/arcface_w600k_r50.hash')
+				}
+			},
+			'sources':
+			{
+				'face_recognizer':
+				{
+					'url': resolve_download_url('models-3.0.0', 'arcface_w600k_r50.onnx'),
+					'path': resolve_relative_path('../.assets/models/arcface_w600k_r50.onnx')
+				}
+			},
+			'template': 'arcface_112_v2',
+			'size': (112, 112)
+		}
+	}
+
+
+def get_inference_pool() -> InferencePool:
+	model_names = [ 'arcface' ]
+	model_source_set = get_model_options().get('sources')
+
+	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)
+
+
+def clear_inference_pool() -> None:
+	model_names = [ 'arcface' ]
+	inference_manager.clear_inference_pool(__name__, model_names)
+
+
+def get_model_options() -> ModelOptions:
+	return create_static_model_set('full').get('arcface')
+
+
+def pre_check() -> bool:
+	model_hash_set = get_model_options().get('hashes')
+	model_source_set = get_model_options().get('sources')
+
+	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)
+
+
+def calc_embedding(temp_vision_frame : VisionFrame, face_landmark_5 : FaceLandmark5) -> Tuple[Embedding, Embedding]:
+	model_template = get_model_options().get('template')
+	model_size = get_model_options().get('size')
+	crop_vision_frame, matrix = warp_face_by_face_landmark_5(temp_vision_frame, face_landmark_5, model_template, model_size)
+	crop_vision_frame = crop_vision_frame / 127.5 - 1
+	crop_vision_frame = crop_vision_frame[:, :, ::-1].transpose(2, 0, 1).astype(numpy.float32)
+	crop_vision_frame = numpy.expand_dims(crop_vision_frame, axis = 0)
+	embedding = forward(crop_vision_frame)
+	embedding = embedding.ravel()
+	normed_embedding = embedding / numpy.linalg.norm(embedding)
+	return embedding, normed_embedding
+
+
+def forward(crop_vision_frame : VisionFrame) -> Embedding:
+	face_recognizer = get_inference_pool().get('face_recognizer')
+
+	with conditional_thread_semaphore():
+		embedding = face_recognizer.run(None,
+		{
+			'input': crop_vision_frame
+		})[0]
+
+	return embedding

facefusion/face_selector.py

@@ -0,0 +1,108 @@
+from typing import List
+
+import numpy
+
+from facefusion import state_manager
+from facefusion.types import Face, FaceSelectorOrder, FaceSet, Gender, Race, Score
+
+
+def find_similar_faces(faces : List[Face], reference_faces : FaceSet, face_distance : float) -> List[Face]:
+	similar_faces : List[Face] = []
+
+	if faces and reference_faces:
+		for reference_set in reference_faces:
+			if not similar_faces:
+				for reference_face in reference_faces[reference_set]:
+					for face in faces:
+						if compare_faces(face, reference_face, face_distance):
+							similar_faces.append(face)
+	return similar_faces
+
+
+def compare_faces(face : Face, reference_face : Face, face_distance : float) -> bool:
+	current_face_distance = calc_face_distance(face, reference_face)
+	current_face_distance = float(numpy.interp(current_face_distance, [ 0, 2 ], [ 0, 1 ]))
+	return current_face_distance < face_distance
+
+
+def calc_face_distance(face : Face, reference_face : Face) -> float:
+	if hasattr(face, 'normed_embedding') and hasattr(reference_face, 'normed_embedding'):
+		return 1 - numpy.dot(face.normed_embedding, reference_face.normed_embedding)
+	return 0
+
+
+def sort_and_filter_faces(faces : List[Face]) -> List[Face]:
+	if faces:
+		if state_manager.get_item('face_selector_order'):
+			faces = sort_faces_by_order(faces, state_manager.get_item('face_selector_order'))
+		if state_manager.get_item('face_selector_gender'):
+			faces = filter_faces_by_gender(faces, state_manager.get_item('face_selector_gender'))
+		if state_manager.get_item('face_selector_race'):
+			faces = filter_faces_by_race(faces, state_manager.get_item('face_selector_race'))
+		if state_manager.get_item('face_selector_age_start') or state_manager.get_item('face_selector_age_end'):
+			faces = filter_faces_by_age(faces, state_manager.get_item('face_selector_age_start'), state_manager.get_item('face_selector_age_end'))
+	return faces
+
+
+def sort_faces_by_order(faces : List[Face], order : FaceSelectorOrder) -> List[Face]:
+	if order == 'left-right':
+		return sorted(faces, key = get_bounding_box_left)
+	if order == 'right-left':
+		return sorted(faces, key = get_bounding_box_left, reverse = True)
+	if order == 'top-bottom':
+		return sorted(faces, key = get_bounding_box_top)
+	if order == 'bottom-top':
+		return sorted(faces, key = get_bounding_box_top, reverse = True)
+	if order == 'small-large':
+		return sorted(faces, key = get_bounding_box_area)
+	if order == 'large-small':
+		return sorted(faces, key = get_bounding_box_area, reverse = True)
+	if order == 'best-worst':
+		return sorted(faces, key = get_face_detector_score, reverse = True)
+	if order == 'worst-best':
+		return sorted(faces, key = get_face_detector_score)
+	return faces
+
+
+def get_bounding_box_left(face : Face) -> float:
+	return face.bounding_box[0]
+
+
+def get_bounding_box_top(face : Face) -> float:
+	return face.bounding_box[1]
+
+
+def get_bounding_box_area(face : Face) -> float:
+	return (face.bounding_box[2] - face.bounding_box[0]) * (face.bounding_box[3] - face.bounding_box[1])
+
+
+def get_face_detector_score(face : Face) -> Score:
+	return face.score_set.get('detector')
+
+
+def filter_faces_by_gender(faces : List[Face], gender : Gender) -> List[Face]:
+	filter_faces = []
+
+	for face in faces:
+		if face.gender == gender:
+			filter_faces.append(face)
+	return filter_faces
+
+
+def filter_faces_by_age(faces : List[Face], face_selector_age_start : int, face_selector_age_end : int) -> List[Face]:
+	filter_faces = []
+	age = range(face_selector_age_start, face_selector_age_end)
+
+	for face in faces:
+		if set(face.age) & set(age):
+			filter_faces.append(face)
+	return filter_faces
+
+
+def filter_faces_by_race(faces : List[Face], race : Race) -> List[Face]:
+	filter_faces = []
+
+	for face in faces:
+		if face.race == race:
+			filter_faces.append(face)
+	return filter_faces

facefusion/face_store.py

@@ -0,0 +1,43 @@
+from typing import List, Optional
+
+from facefusion.hash_helper import create_hash
+from facefusion.types import Face, FaceSet, FaceStore, VisionFrame
+
+FACE_STORE : FaceStore =\
+{
+	'static_faces': {},
+	'reference_faces': {}
+}
+
+
+def get_face_store() -> FaceStore:
+	return FACE_STORE
+
+
+def get_static_faces(vision_frame : VisionFrame) -> Optional[List[Face]]:
+	vision_hash = create_hash(vision_frame.tobytes())
+	return FACE_STORE.get('static_faces').get(vision_hash)
+
+
+def set_static_faces(vision_frame : VisionFrame, faces : List[Face]) -> None:
+	vision_hash = create_hash(vision_frame.tobytes())
+	if vision_hash:
+		FACE_STORE['static_faces'][vision_hash] = faces
+
+
+def clear_static_faces() -> None:
+	FACE_STORE['static_faces'].clear()
+
+
+def get_reference_faces() -> Optional[FaceSet]:
+	return FACE_STORE.get('reference_faces')
+
+
+def append_reference_face(name : str, face : Face) -> None:
+	if name not in FACE_STORE.get('reference_faces'):
+		FACE_STORE['reference_faces'][name] = []
+	FACE_STORE['reference_faces'][name].append(face)
+
+
+def clear_reference_faces() -> None:
+	FACE_STORE['reference_faces'].clear()

facefusion/ffmpeg.py

@@ -0,0 +1,286 @@
+import os
+import subprocess
+import tempfile
+from functools import partial
+from typing import List, Optional, cast
+
+from tqdm import tqdm
+
+import facefusion.choices
+from facefusion import ffmpeg_builder, logger, process_manager, state_manager, wording
+from facefusion.filesystem import get_file_format, remove_file
+from facefusion.temp_helper import get_temp_file_path, get_temp_frames_pattern
+from facefusion.types import AudioBuffer, AudioEncoder, Commands, EncoderSet, Fps, UpdateProgress, VideoEncoder, VideoFormat
+from facefusion.vision import detect_video_duration, detect_video_fps, predict_video_frame_total
+
+
+def run_ffmpeg_with_progress(commands : Commands, update_progress : UpdateProgress) -> subprocess.Popen[bytes]:
+	log_level = state_manager.get_item('log_level')
+	commands.extend(ffmpeg_builder.set_progress())
+	commands.extend(ffmpeg_builder.cast_stream())
+	commands = ffmpeg_builder.run(commands)
+	process = subprocess.Popen(commands, stderr = subprocess.PIPE, stdout = subprocess.PIPE)
+
+	while process_manager.is_processing():
+		try:
+
+			while __line__ := process.stdout.readline().decode().lower():
+				if 'frame=' in __line__:
+					_, frame_number = __line__.split('frame=')
+					update_progress(int(frame_number))
+
+			if log_level == 'debug':
+				log_debug(process)
+			process.wait(timeout = 0.5)
+		except subprocess.TimeoutExpired:
+			continue
+		return process
+
+	if process_manager.is_stopping():
+		process.terminate()
+	return process
+
+
+def update_progress(progress : tqdm, frame_number : int) -> None:
+	progress.update(frame_number - progress.n)
+
+
+def run_ffmpeg(commands : Commands) -> subprocess.Popen[bytes]:
+	log_level = state_manager.get_item('log_level')
+	commands = ffmpeg_builder.run(commands)
+	process = subprocess.Popen(commands, stderr = subprocess.PIPE, stdout = subprocess.PIPE)
+
+	while process_manager.is_processing():
+		try:
+			if log_level == 'debug':
+				log_debug(process)
+			process.wait(timeout = 0.5)
+		except subprocess.TimeoutExpired:
+			continue
+		return process
+
+	if process_manager.is_stopping():
+		process.terminate()
+	return process
+
+
+def open_ffmpeg(commands : Commands) -> subprocess.Popen[bytes]:
+	commands = ffmpeg_builder.run(commands)
+	return subprocess.Popen(commands, stdin = subprocess.PIPE, stdout = subprocess.PIPE)
+
+
+def log_debug(process : subprocess.Popen[bytes]) -> None:
+	_, stderr = process.communicate()
+	errors = stderr.decode().split(os.linesep)
+
+	for error in errors:
+		if error.strip():
+			logger.debug(error.strip(), __name__)
+
+
+def get_available_encoder_set() -> EncoderSet:
+	available_encoder_set : EncoderSet =\
+	{
+		'audio': [],
+		'video': []
+	}
+	commands = ffmpeg_builder.chain(
+		ffmpeg_builder.get_encoders()
+	)
+	process = run_ffmpeg(commands)
+
+	while line := process.stdout.readline().decode().lower():
+		if line.startswith(' a'):
+			audio_encoder = line.split()[1]
+
+			if audio_encoder in facefusion.choices.output_audio_encoders:
+				index = facefusion.choices.output_audio_encoders.index(audio_encoder) #type:ignore[arg-type]
+				available_encoder_set['audio'].insert(index, audio_encoder) #type:ignore[arg-type]
+		if line.startswith(' v'):
+			video_encoder = line.split()[1]
+
+			if video_encoder in facefusion.choices.output_video_encoders:
+				index = facefusion.choices.output_video_encoders.index(video_encoder) #type:ignore[arg-type]
+				available_encoder_set['video'].insert(index, video_encoder) #type:ignore[arg-type]
+
+	return available_encoder_set
+
+
+def extract_frames(target_path : str, temp_video_resolution : str, temp_video_fps : Fps, trim_frame_start : int, trim_frame_end : int) -> bool:
+	extract_frame_total = predict_video_frame_total(target_path, temp_video_fps, trim_frame_start, trim_frame_end)
+	temp_frames_pattern = get_temp_frames_pattern(target_path, '%08d')
+	commands = ffmpeg_builder.chain(
+		ffmpeg_builder.set_input(target_path),
+		ffmpeg_builder.set_media_resolution(temp_video_resolution),
+		ffmpeg_builder.set_frame_quality(0),
+		ffmpeg_builder.select_frame_range(trim_frame_start, trim_frame_end, temp_video_fps),
+		ffmpeg_builder.prevent_frame_drop(),
+		ffmpeg_builder.set_output(temp_frames_pattern)
+	)
+
+	with tqdm(total = extract_frame_total, desc = wording.get('extracting'), unit = 'frame', ascii = ' =', disable = state_manager.get_item('log_level') in [ 'warn', 'error' ]) as progress:
+		process = run_ffmpeg_with_progress(commands, partial(update_progress, progress))
+		return process.returncode == 0
+
+
+def copy_image(target_path : str, temp_image_resolution : str) -> bool:
+	temp_image_path = get_temp_file_path(target_path)
+	commands = ffmpeg_builder.chain(
+		ffmpeg_builder.set_input(target_path),
+		ffmpeg_builder.set_media_resolution(temp_image_resolution),
+		ffmpeg_builder.set_image_quality(target_path, 100),
+		ffmpeg_builder.force_output(temp_image_path)
+	)
+	return run_ffmpeg(commands).returncode == 0
+
+
+def finalize_image(target_path : str, output_path : str, output_image_resolution : str) -> bool:
+	output_image_quality = state_manager.get_item('output_image_quality')
+	temp_image_path = get_temp_file_path(target_path)
+	commands = ffmpeg_builder.chain(
+		ffmpeg_builder.set_input(temp_image_path),
+		ffmpeg_builder.set_media_resolution(output_image_resolution),
+		ffmpeg_builder.set_image_quality(target_path, output_image_quality),
+		ffmpeg_builder.force_output(output_path)
+	)
+	return run_ffmpeg(commands).returncode == 0
+
+
+def read_audio_buffer(target_path : str, audio_sample_rate : int, audio_sample_size : int, audio_channel_total : int) -> Optional[AudioBuffer]:
+	commands = ffmpeg_builder.chain(
+		ffmpeg_builder.set_input(target_path),
+		ffmpeg_builder.ignore_video_stream(),
+		ffmpeg_builder.set_audio_sample_rate(audio_sample_rate),
+		ffmpeg_builder.set_audio_sample_size(audio_sample_size),
+		ffmpeg_builder.set_audio_channel_total(audio_channel_total),
+		ffmpeg_builder.cast_stream()
+	)
+
+	process = open_ffmpeg(commands)
+	audio_buffer, _ = process.communicate()
+	if process.returncode == 0:
+		return audio_buffer
+	return None
+
+
+def restore_audio(target_path : str, output_path : str, trim_frame_start : int, trim_frame_end : int) -> bool:
+	output_audio_encoder = state_manager.get_item('output_audio_encoder')
+	output_audio_quality = state_manager.get_item('output_audio_quality')
+	output_audio_volume = state_manager.get_item('output_audio_volume')
+	target_video_fps = detect_video_fps(target_path)
+	temp_video_path = get_temp_file_path(target_path)
+	temp_video_format = cast(VideoFormat, get_file_format(temp_video_path))
+	temp_video_duration = detect_video_duration(temp_video_path)
+
+	output_audio_encoder = fix_audio_encoder(temp_video_format, output_audio_encoder)
+	commands = ffmpeg_builder.chain(
+		ffmpeg_builder.set_input(temp_video_path),
+		ffmpeg_builder.select_media_range(trim_frame_start, trim_frame_end, target_video_fps),
+		ffmpeg_builder.set_input(target_path),
+		ffmpeg_builder.copy_video_encoder(),
+		ffmpeg_builder.set_audio_encoder(output_audio_encoder),
+		ffmpeg_builder.set_audio_quality(output_audio_encoder, output_audio_quality),
+		ffmpeg_builder.set_audio_volume(output_audio_volume),
+		ffmpeg_builder.select_media_stream('0:v:0'),
+		ffmpeg_builder.select_media_stream('1:a:0'),
+		ffmpeg_builder.set_video_duration(temp_video_duration),
+		ffmpeg_builder.force_output(output_path)
+	)
+	return run_ffmpeg(commands).returncode == 0
+
+
+def replace_audio(target_path : str, audio_path : str, output_path : str) -> bool:
+	output_audio_encoder = state_manager.get_item('output_audio_encoder')
+	output_audio_quality = state_manager.get_item('output_audio_quality')
+	output_audio_volume = state_manager.get_item('output_audio_volume')
+	temp_video_path = get_temp_file_path(target_path)
+	temp_video_format = cast(VideoFormat, get_file_format(temp_video_path))
+	temp_video_duration = detect_video_duration(temp_video_path)
+
+	output_audio_encoder = fix_audio_encoder(temp_video_format, output_audio_encoder)
+	commands = ffmpeg_builder.chain(
+		ffmpeg_builder.set_input(temp_video_path),
+		ffmpeg_builder.set_input(audio_path),
+		ffmpeg_builder.copy_video_encoder(),
+		ffmpeg_builder.set_audio_encoder(output_audio_encoder),
+		ffmpeg_builder.set_audio_quality(output_audio_encoder, output_audio_quality),
+		ffmpeg_builder.set_audio_volume(output_audio_volume),
+		ffmpeg_builder.set_video_duration(temp_video_duration),
+		ffmpeg_builder.force_output(output_path)
+	)
+	return run_ffmpeg(commands).returncode == 0
+
+
+def merge_video(target_path : str, temp_video_fps : Fps, output_video_resolution : str, output_video_fps : Fps, trim_frame_start : int, trim_frame_end : int) -> bool:
+	output_video_encoder = state_manager.get_item('output_video_encoder')
+	output_video_quality = state_manager.get_item('output_video_quality')
+	output_video_preset = state_manager.get_item('output_video_preset')
+	merge_frame_total = predict_video_frame_total(target_path, output_video_fps, trim_frame_start, trim_frame_end)
+	temp_video_path = get_temp_file_path(target_path)
+	temp_video_format = cast(VideoFormat, get_file_format(temp_video_path))
+	temp_frames_pattern = get_temp_frames_pattern(target_path, '%08d')
+
+	output_video_encoder = fix_video_encoder(temp_video_format, output_video_encoder)
+	commands = ffmpeg_builder.chain(
+		ffmpeg_builder.set_input_fps(temp_video_fps),
+		ffmpeg_builder.set_input(temp_frames_pattern),
+		ffmpeg_builder.set_media_resolution(output_video_resolution),
+		ffmpeg_builder.set_video_encoder(output_video_encoder),
+		ffmpeg_builder.set_video_quality(output_video_encoder, output_video_quality),
+		ffmpeg_builder.set_video_preset(output_video_encoder, output_video_preset),
+		ffmpeg_builder.set_video_fps(output_video_fps),
+		ffmpeg_builder.set_pixel_format(output_video_encoder),
+		ffmpeg_builder.set_video_colorspace('bt709'),
+		ffmpeg_builder.force_output(temp_video_path)
+	)
+
+	with tqdm(total = merge_frame_total, desc = wording.get('merging'), unit = 'frame', ascii = ' =', disable = state_manager.get_item('log_level') in [ 'warn', 'error' ]) as progress:
+		process = run_ffmpeg_with_progress(commands, partial(update_progress, progress))
+		return process.returncode == 0
+
+
+def concat_video(output_path : str, temp_output_paths : List[str]) -> bool:
+	concat_video_path = tempfile.mktemp()
+
+	with open(concat_video_path, 'w') as concat_video_file:
+		for temp_output_path in temp_output_paths:
+			concat_video_file.write('file \'' + os.path.abspath(temp_output_path) + '\'' + os.linesep)
+		concat_video_file.flush()
+		concat_video_file.close()
+
+	output_path = os.path.abspath(output_path)
+	commands = ffmpeg_builder.chain(
+		ffmpeg_builder.unsafe_concat(),
+		ffmpeg_builder.set_input(concat_video_file.name),
+		ffmpeg_builder.copy_video_encoder(),
+		ffmpeg_builder.copy_audio_encoder(),
+		ffmpeg_builder.force_output(output_path)
+	)
+	process = run_ffmpeg(commands)
+	process.communicate()
+	remove_file(concat_video_path)
+	return process.returncode == 0
+
+
+def fix_audio_encoder(video_format : VideoFormat, audio_encoder : AudioEncoder) -> AudioEncoder:
+	if video_format == 'avi' and audio_encoder == 'libopus':
+		return 'aac'
+	if video_format == 'm4v':
+		return 'aac'
+	if video_format == 'mov' and audio_encoder in [ 'flac', 'libopus' ]:
+		return 'aac'
+	if video_format == 'webm':
+		return 'libopus'
+	return audio_encoder
+
+
+def fix_video_encoder(video_format : VideoFormat, video_encoder : VideoEncoder) -> VideoEncoder:
+	if video_format == 'm4v':
+		return 'libx264'
+	if video_format in [ 'mkv', 'mp4' ] and video_encoder == 'rawvideo':
+		return 'libx264'
+	if video_format == 'mov' and video_encoder == 'libvpx-vp9':
+		return 'libx264'
+	if video_format == 'webm':
+		return 'libvpx-vp9'
+	return video_encoder

facefusion/ffmpeg_builder.py

@@ -0,0 +1,248 @@
+import itertools
+import shutil
+from typing import Optional
+
+import numpy
+
+from facefusion.filesystem import get_file_format
+from facefusion.types import AudioEncoder, Commands, Duration, Fps, StreamMode, VideoEncoder, VideoPreset
+
+
+def run(commands : Commands) -> Commands:
+	return [ shutil.which('ffmpeg'), '-loglevel', 'error' ] + commands
+
+
+def chain(*commands : Commands) -> Commands:
+	return list(itertools.chain(*commands))
+
+
+def get_encoders() -> Commands:
+	return [ '-encoders' ]
+
+
+def set_hardware_accelerator(value : str) -> Commands:
+	return [ '-hwaccel', value ]
+
+
+def set_progress() -> Commands:
+	return [ '-progress' ]
+
+
+def set_input(input_path : str) -> Commands:
+	return [ '-i', input_path ]
+
+
+def set_input_fps(input_fps : Fps) -> Commands:
+	return [ '-r', str(input_fps)]
+
+
+def set_output(output_path : str) -> Commands:
+	return [ output_path ]
+
+
+def force_output(output_path : str) -> Commands:
+	return [ '-y', output_path ]
+
+
+def cast_stream() -> Commands:
+	return [ '-' ]
+
+
+def set_stream_mode(stream_mode : StreamMode) -> Commands:
+	if stream_mode == 'udp':
+		return [ '-f', 'mpegts' ]
+	if stream_mode == 'v4l2':
+		return [ '-f', 'v4l2' ]
+	return []
+
+
+def set_stream_quality(stream_quality : int) -> Commands:
+	return [ '-b:v', str(stream_quality) + 'k' ]
+
+
+def unsafe_concat() -> Commands:
+	return [ '-f', 'concat', '-safe', '0' ]
+
+
+def set_pixel_format(video_encoder : VideoEncoder) -> Commands:
+	if video_encoder == 'rawvideo':
+		return [ '-pix_fmt', 'rgb24' ]
+	return [ '-pix_fmt', 'yuv420p' ]
+
+
+def set_frame_quality(frame_quality : int) -> Commands:
+	return [ '-q:v', str(frame_quality) ]
+
+
+def select_frame_range(frame_start : int, frame_end : int, video_fps : Fps) -> Commands:
+	if isinstance(frame_start, int) and isinstance(frame_end, int):
+		return [ '-vf', 'trim=start_frame=' + str(frame_start) + ':end_frame=' + str(frame_end) + ',fps=' + str(video_fps) ]
+	if isinstance(frame_start, int):
+		return [ '-vf', 'trim=start_frame=' + str(frame_start) + ',fps=' + str(video_fps) ]
+	if isinstance(frame_end, int):
+		return [ '-vf', 'trim=end_frame=' + str(frame_end) + ',fps=' + str(video_fps) ]
+	return [ '-vf', 'fps=' + str(video_fps) ]
+
+
+def prevent_frame_drop() -> Commands:
+	return [ '-vsync', '0' ]
+
+
+def select_media_range(frame_start : int, frame_end : int, media_fps : Fps) -> Commands:
+	commands = []
+
+	if isinstance(frame_start, int):
+		commands.extend([ '-ss', str(frame_start / media_fps) ])
+	if isinstance(frame_end, int):
+		commands.extend([ '-to', str(frame_end / media_fps) ])
+	return commands
+
+
+def select_media_stream(media_stream : str) -> Commands:
+	return [ '-map', media_stream ]
+
+
+def set_media_resolution(video_resolution : str) -> Commands:
+	return [ '-s', video_resolution ]
+
+
+def set_image_quality(image_path : str, image_quality : int) -> Commands:
+	if get_file_format(image_path) == 'webp':
+		image_compression = image_quality
+	else:
+		image_compression = round(31 - (image_quality * 0.31))
+	return [ '-q:v', str(image_compression) ]
+
+
+def set_audio_encoder(audio_codec : str) -> Commands:
+	return [ '-c:a', audio_codec ]
+
+
+def copy_audio_encoder() -> Commands:
+	return set_audio_encoder('copy')
+
+
+def set_audio_sample_rate(audio_sample_rate : int) -> Commands:
+	return [ '-ar', str(audio_sample_rate) ]
+
+
+def set_audio_sample_size(audio_sample_size : int) -> Commands:
+	if audio_sample_size == 16:
+		return [ '-f', 's16le' ]
+	if audio_sample_size == 32:
+		return [ '-f', 's32le' ]
+	return []
+
+
+def set_audio_channel_total(audio_channel_total : int) -> Commands:
+	return [ '-ac', str(audio_channel_total) ]
+
+
+def set_audio_quality(audio_encoder : AudioEncoder, audio_quality : int) -> Commands:
+	if audio_encoder == 'aac':
+		audio_compression = round(numpy.interp(audio_quality, [ 0, 100 ], [ 0.1, 2.0 ]), 1)
+		return [ '-q:a', str(audio_compression) ]
+	if audio_encoder == 'libmp3lame':
+		audio_compression = round(numpy.interp(audio_quality, [ 0, 100 ], [ 9, 0 ]))
+		return [ '-q:a', str(audio_compression) ]
+	if audio_encoder == 'libopus':
+		audio_bit_rate = round(numpy.interp(audio_quality, [ 0, 100 ], [ 64, 256 ]))
+		return [ '-b:a', str(audio_bit_rate) + 'k' ]
+	if audio_encoder == 'libvorbis':
+		audio_compression = round(numpy.interp(audio_quality, [ 0, 100 ], [ -1, 10 ]), 1)
+		return [ '-q:a', str(audio_compression) ]
+	return []
+
+
+def set_audio_volume(audio_volume : int) -> Commands:
+	return [ '-filter:a', 'volume=' + str(audio_volume / 100) ]
+
+
+def set_video_encoder(video_encoder : str) -> Commands:
+	return [ '-c:v', video_encoder ]
+
+
+def copy_video_encoder() -> Commands:
+	return set_video_encoder('copy')
+
+
+def set_video_quality(video_encoder : VideoEncoder, video_quality : int) -> Commands:
+	if video_encoder in [ 'libx264', 'libx265' ]:
+		video_compression = round(numpy.interp(video_quality, [ 0, 100 ], [ 51, 0 ]))
+		return [ '-crf', str(video_compression) ]
+	if video_encoder == 'libvpx-vp9':
+		video_compression = round(numpy.interp(video_quality, [ 0, 100 ], [ 63, 0 ]))
+		return [ '-crf', str(video_compression) ]
+	if video_encoder in [ 'h264_nvenc', 'hevc_nvenc' ]:
+		video_compression = round(numpy.interp(video_quality, [ 0, 100 ], [ 51, 0 ]))
+		return [ '-cq', str(video_compression) ]
+	if video_encoder in [ 'h264_amf', 'hevc_amf' ]:
+		video_compression = round(numpy.interp(video_quality, [ 0, 100 ], [ 51, 0 ]))
+		return [ '-qp_i', str(video_compression), '-qp_p', str(video_compression), '-qp_b', str(video_compression) ]
+	if video_encoder in [ 'h264_qsv', 'hevc_qsv' ]:
+		video_compression = round(numpy.interp(video_quality, [ 0, 100 ], [ 51, 0 ]))
+		return [ '-qp', str(video_compression) ]
+	if video_encoder in [ 'h264_videotoolbox', 'hevc_videotoolbox' ]:
+		video_bit_rate = round(numpy.interp(video_quality, [ 0, 100 ], [ 1024, 50512 ]))
+		return [ '-b:v', str(video_bit_rate) + 'k' ]
+	return []
+
+
+def set_video_preset(video_encoder : VideoEncoder, video_preset : VideoPreset) -> Commands:
+	if video_encoder in [ 'libx264', 'libx265' ]:
+		return [ '-preset', video_preset ]
+	if video_encoder in [ 'h264_nvenc', 'hevc_nvenc' ]:
+		return [ '-preset', map_nvenc_preset(video_preset) ]
+	if video_encoder in [ 'h264_amf', 'hevc_amf' ]:
+		return [ '-quality', map_amf_preset(video_preset) ]
+	if video_encoder in [ 'h264_qsv', 'hevc_qsv' ]:
+		return [ '-preset', map_qsv_preset(video_preset) ]
+	return []
+
+
+def set_video_colorspace(video_colorspace : str) -> Commands:
+	return [ '-colorspace', video_colorspace ]
+
+
+def set_video_fps(video_fps : Fps) -> Commands:
+	return [ '-vf', 'framerate=fps=' + str(video_fps) ]
+
+
+def set_video_duration(video_duration : Duration) -> Commands:
+	return [ '-t', str(video_duration) ]
+
+
+def capture_video() -> Commands:
+	return [ '-f', 'rawvideo', '-pix_fmt', 'rgb24' ]
+
+
+def ignore_video_stream() -> Commands:
+	return [ '-vn' ]
+
+
+def map_nvenc_preset(video_preset : VideoPreset) -> Optional[str]:
+	if video_preset in [ 'ultrafast', 'superfast', 'veryfast', 'faster', 'fast' ]:
+		return 'fast'
+	if video_preset == 'medium':
+		return 'medium'
+	if video_preset in [ 'slow', 'slower', 'veryslow' ]:
+		return 'slow'
+	return None
+
+
+def map_amf_preset(video_preset : VideoPreset) -> Optional[str]:
+	if video_preset in [ 'ultrafast', 'superfast', 'veryfast' ]:
+		return 'speed'
+	if video_preset in [ 'faster', 'fast', 'medium' ]:
+		return 'balanced'
+	if video_preset in [ 'slow', 'slower', 'veryslow' ]:
+		return 'quality'
+	return None
+
+
+def map_qsv_preset(video_preset : VideoPreset) -> Optional[str]:
+	if video_preset in [ 'ultrafast', 'superfast', 'veryfast' ]:
+		return 'veryfast'
+	if video_preset in [ 'faster', 'fast', 'medium', 'slow', 'slower', 'veryslow' ]:
+		return video_preset
+	return None

facefusion/filesystem.py

@@ -0,0 +1,188 @@
+import glob
+import os
+import shutil
+from typing import List, Optional
+
+import facefusion.choices
+
+
+def get_file_size(file_path : str) -> int:
+	if is_file(file_path):
+		return os.path.getsize(file_path)
+	return 0
+
+
+def get_file_name(file_path : str) -> Optional[str]:
+	file_name, _ = os.path.splitext(os.path.basename(file_path))
+
+	if file_name:
+		return file_name
+	return None
+
+
+def get_file_extension(file_path : str) -> Optional[str]:
+	_, file_extension = os.path.splitext(file_path)
+
+	if file_extension:
+		return file_extension.lower()
+	return None
+
+
+def get_file_format(file_path : str) -> Optional[str]:
+	file_extension = get_file_extension(file_path)
+
+	if file_extension:
+		if file_extension == '.jpg':
+			return 'jpeg'
+		if file_extension == '.tif':
+			return 'tiff'
+		return file_extension.lstrip('.')
+	return None
+
+
+def same_file_extension(first_file_path : str, second_file_path : str) -> bool:
+	first_file_extension = get_file_extension(first_file_path)
+	second_file_extension = get_file_extension(second_file_path)
+
+	if first_file_extension and second_file_extension:
+		return get_file_extension(first_file_path) == get_file_extension(second_file_path)
+	return False
+
+
+def is_file(file_path : str) -> bool:
+	if file_path:
+		return os.path.isfile(file_path)
+	return False
+
+
+def is_audio(audio_path : str) -> bool:
+	return is_file(audio_path) and get_file_format(audio_path) in facefusion.choices.audio_formats
+
+
+def has_audio(audio_paths : List[str]) -> bool:
+	if audio_paths:
+		return any(map(is_audio, audio_paths))
+	return False
+
+
+def are_audios(audio_paths : List[str]) -> bool:
+	if audio_paths:
+		return all(map(is_audio, audio_paths))
+	return False
+
+
+def is_image(image_path : str) -> bool:
+	return is_file(image_path) and get_file_format(image_path) in facefusion.choices.image_formats
+
+
+def has_image(image_paths : List[str]) -> bool:
+	if image_paths:
+		return any(is_image(image_path) for image_path in image_paths)
+	return False
+
+
+def are_images(image_paths : List[str]) -> bool:
+	if image_paths:
+		return all(map(is_image, image_paths))
+	return False
+
+
+def is_video(video_path : str) -> bool:
+	return is_file(video_path) and get_file_format(video_path) in facefusion.choices.video_formats
+
+
+def has_video(video_paths : List[str]) -> bool:
+	if video_paths:
+		return any(map(is_video, video_paths))
+	return False
+
+
+def are_videos(video_paths : List[str]) -> bool:
+	if video_paths:
+		return any(map(is_video, video_paths))
+	return False
+
+
+def filter_audio_paths(paths : List[str]) -> List[str]:
+	if paths:
+		return [ path for path in paths if is_audio(path) ]
+	return []
+
+
+def filter_image_paths(paths : List[str]) -> List[str]:
+	if paths:
+		return [ path for path in paths if is_image(path) ]
+	return []
+
+
+def copy_file(file_path : str, move_path : str) -> bool:
+	if is_file(file_path):
+		shutil.copy(file_path, move_path)
+		return is_file(move_path)
+	return False
+
+
+def move_file(file_path : str, move_path : str) -> bool:
+	if is_file(file_path):
+		shutil.move(file_path, move_path)
+		return not is_file(file_path) and is_file(move_path)
+	return False
+
+
+def remove_file(file_path : str) -> bool:
+	if is_file(file_path):
+		os.remove(file_path)
+		return not is_file(file_path)
+	return False
+
+
+def resolve_file_paths(directory_path : str) -> List[str]:
+	file_paths : List[str] = []
+
+	if is_directory(directory_path):
+		file_names_and_extensions = sorted(os.listdir(directory_path))
+
+		for file_name_and_extension in file_names_and_extensions:
+			if not file_name_and_extension.startswith(('.', '__')):
+				file_path = os.path.join(directory_path, file_name_and_extension)
+				file_paths.append(file_path)
+
+	return file_paths
+
+
+def resolve_file_pattern(file_pattern : str) -> List[str]:
+	if in_directory(file_pattern):
+		return sorted(glob.glob(file_pattern))
+	return []
+
+
+def is_directory(directory_path : str) -> bool:
+	if directory_path:
+		return os.path.isdir(directory_path)
+	return False
+
+
+def in_directory(file_path : str) -> bool:
+	if file_path:
+		directory_path = os.path.dirname(file_path)
+		if directory_path:
+			return not is_directory(file_path) and is_directory(directory_path)
+	return False
+
+
+def create_directory(directory_path : str) -> bool:
+	if directory_path and not is_file(directory_path):
+		os.makedirs(directory_path, exist_ok = True)
+		return is_directory(directory_path)
+	return False
+
+
+def remove_directory(directory_path : str) -> bool:
+	if is_directory(directory_path):
+		shutil.rmtree(directory_path, ignore_errors = True)
+		return not is_directory(directory_path)
+	return False
+
+
+def resolve_relative_path(path : str) -> str:
+	return os.path.abspath(os.path.join(os.path.dirname(__file__), path))

facefusion/hash_helper.py

@@ -0,0 +1,32 @@
+import os
+import zlib
+from typing import Optional
+
+from facefusion.filesystem import get_file_name, is_file
+
+
+def create_hash(content : bytes) -> str:
+	return format(zlib.crc32(content), '08x')
+
+
+def validate_hash(validate_path : str) -> bool:
+	hash_path = get_hash_path(validate_path)
+
+	if is_file(hash_path):
+		with open(hash_path) as hash_file:
+			hash_content = hash_file.read()
+
+		with open(validate_path, 'rb') as validate_file:
+			validate_content = validate_file.read()
+
+		return create_hash(validate_content) == hash_content
+	return False
+
+
+def get_hash_path(validate_path : str) -> Optional[str]:
+	if is_file(validate_path):
+		validate_directory_path, file_name_and_extension = os.path.split(validate_path)
+		validate_file_name = get_file_name(file_name_and_extension)
+
+		return os.path.join(validate_directory_path, validate_file_name + '.hash')
+	return None

facefusion/inference_manager.py

@@ -0,0 +1,74 @@
+import importlib
+from time import sleep
+from typing import List
+
+from onnxruntime import InferenceSession
+
+from facefusion import process_manager, state_manager
+from facefusion.app_context import detect_app_context
+from facefusion.execution import create_inference_session_providers
+from facefusion.filesystem import is_file
+from facefusion.types import DownloadSet, ExecutionProvider, InferencePool, InferencePoolSet
+
+INFERENCE_POOL_SET : InferencePoolSet =\
+{
+	'cli': {},
+	'ui': {}
+}
+
+
+def get_inference_pool(module_name : str, model_names : List[str], model_source_set : DownloadSet) -> InferencePool:
+	while process_manager.is_checking():
+		sleep(0.5)
+	execution_device_id = state_manager.get_item('execution_device_id')
+	execution_providers = resolve_execution_providers(module_name)
+	app_context = detect_app_context()
+	inference_context = get_inference_context(module_name, model_names, execution_device_id, execution_providers)
+
+	if app_context == 'cli' and INFERENCE_POOL_SET.get('ui').get(inference_context):
+		INFERENCE_POOL_SET['cli'][inference_context] = INFERENCE_POOL_SET.get('ui').get(inference_context)
+	if app_context == 'ui' and INFERENCE_POOL_SET.get('cli').get(inference_context):
+		INFERENCE_POOL_SET['ui'][inference_context] = INFERENCE_POOL_SET.get('cli').get(inference_context)
+	if not INFERENCE_POOL_SET.get(app_context).get(inference_context):
+		INFERENCE_POOL_SET[app_context][inference_context] = create_inference_pool(model_source_set, execution_device_id, execution_providers)
+
+	return INFERENCE_POOL_SET.get(app_context).get(inference_context)
+
+
+def create_inference_pool(model_source_set : DownloadSet, execution_device_id : str, execution_providers : List[ExecutionProvider]) -> InferencePool:
+	inference_pool : InferencePool = {}
+
+	for model_name in model_source_set.keys():
+		model_path = model_source_set.get(model_name).get('path')
+		if is_file(model_path):
+			inference_pool[model_name] = create_inference_session(model_path, execution_device_id, execution_providers)
+
+	return inference_pool
+
+
+def clear_inference_pool(module_name : str, model_names : List[str]) -> None:
+	execution_device_id = state_manager.get_item('execution_device_id')
+	execution_providers = resolve_execution_providers(module_name)
+	app_context = detect_app_context()
+	inference_context = get_inference_context(module_name, model_names, execution_device_id, execution_providers)
+
+	if INFERENCE_POOL_SET.get(app_context).get(inference_context):
+		del INFERENCE_POOL_SET[app_context][inference_context]
+
+
+def create_inference_session(model_path : str, execution_device_id : str, execution_providers : List[ExecutionProvider]) -> InferenceSession:
+	inference_session_providers = create_inference_session_providers(execution_device_id, execution_providers)
+	return InferenceSession(model_path, providers = inference_session_providers)
+
+
+def get_inference_context(module_name : str, model_names : List[str], execution_device_id : str, execution_providers : List[ExecutionProvider]) -> str:
+	inference_context = '.'.join([ module_name ] + model_names + [ execution_device_id ] + list(execution_providers))
+	return inference_context
+
+
+def resolve_execution_providers(module_name : str) -> List[ExecutionProvider]:
+	module = importlib.import_module(module_name)
+
+	if hasattr(module, 'resolve_execution_providers'):
+		return getattr(module, 'resolve_execution_providers')()
+	return state_manager.get_item('execution_providers')

facefusion/installer.py

@@ -0,0 +1,96 @@
+import os
+import shutil
+import signal
+import subprocess
+import sys
+from argparse import ArgumentParser, HelpFormatter
+from functools import partial
+from types import FrameType
+
+from facefusion import metadata, wording
+from facefusion.common_helper import is_linux, is_windows
+
+ONNXRUNTIME_SET =\
+{
+	'default': ('onnxruntime', '1.22.0')
+}
+if is_windows() or is_linux():
+	ONNXRUNTIME_SET['cuda'] = ('onnxruntime-gpu', '1.22.0')
+	ONNXRUNTIME_SET['openvino'] = ('onnxruntime-openvino', '1.22.0')
+if is_windows():
+	ONNXRUNTIME_SET['directml'] = ('onnxruntime-directml', '1.17.3')
+if is_linux():
+	ONNXRUNTIME_SET['rocm'] = ('onnxruntime-rocm', '1.21.0')
+
+
+def cli() -> None:
+	signal.signal(signal.SIGINT, signal_exit)
+	program = ArgumentParser(formatter_class = partial(HelpFormatter, max_help_position = 50))
+	program.add_argument('--onnxruntime', help = wording.get('help.install_dependency').format(dependency = 'onnxruntime'), choices = ONNXRUNTIME_SET.keys(), required = True)
+	program.add_argument('--skip-conda', help = wording.get('help.skip_conda'), action = 'store_true')
+	program.add_argument('-v', '--version', version = metadata.get('name') + ' ' + metadata.get('version'), action = 'version')
+	run(program)
+
+
+def signal_exit(signum : int, frame : FrameType) -> None:
+	sys.exit(0)
+
+
+def run(program : ArgumentParser) -> None:
+	args = program.parse_args()
+	has_conda = 'CONDA_PREFIX' in os.environ
+	onnxruntime_name, onnxruntime_version = ONNXRUNTIME_SET.get(args.onnxruntime)
+
+	if not args.skip_conda and not has_conda:
+		sys.stdout.write(wording.get('conda_not_activated') + os.linesep)
+		sys.exit(1)
+
+	with open('requirements.txt') as file:
+
+		for line in file.readlines():
+			__line__ = line.strip()
+			if not __line__.startswith('onnxruntime'):
+				subprocess.call([ shutil.which('pip'), 'install', line, '--force-reinstall' ])
+
+	if args.onnxruntime == 'rocm':
+		python_id = 'cp' + str(sys.version_info.major) + str(sys.version_info.minor)
+
+		if python_id in [ 'cp310', 'cp312' ]:
+			wheel_name = 'onnxruntime_rocm-' + onnxruntime_version + '-' + python_id + '-' + python_id + '-linux_x86_64.whl'
+			wheel_url = 'https://repo.radeon.com/rocm/manylinux/rocm-rel-6.4/' + wheel_name
+			subprocess.call([ shutil.which('pip'), 'install', wheel_url, '--force-reinstall' ])
+	else:
+		subprocess.call([ shutil.which('pip'), 'install', onnxruntime_name + '==' + onnxruntime_version, '--force-reinstall' ])
+
+	if args.onnxruntime == 'cuda' and has_conda:
+		library_paths = []
+
+		if is_linux():
+			if os.getenv('LD_LIBRARY_PATH'):
+				library_paths = os.getenv('LD_LIBRARY_PATH').split(os.pathsep)
+
+			python_id = 'python' + str(sys.version_info.major) + '.' + str(sys.version_info.minor)
+			library_paths.extend(
+			[
+				os.path.join(os.getenv('CONDA_PREFIX'), 'lib'),
+				os.path.join(os.getenv('CONDA_PREFIX'), 'lib', python_id, 'site-packages', 'tensorrt_libs')
+			])
+			library_paths = list(dict.fromkeys([ library_path for library_path in library_paths if os.path.exists(library_path) ]))
+
+			subprocess.call([ shutil.which('conda'), 'env', 'config', 'vars', 'set', 'LD_LIBRARY_PATH=' + os.pathsep.join(library_paths) ])
+
+		if is_windows():
+			if os.getenv('PATH'):
+				library_paths = os.getenv('PATH').split(os.pathsep)
+
+			library_paths.extend(
+			[
+				os.path.join(os.getenv('CONDA_PREFIX'), 'Lib'),
+				os.path.join(os.getenv('CONDA_PREFIX'), 'Lib', 'site-packages', 'tensorrt_libs')
+			])
+			library_paths = list(dict.fromkeys([ library_path for library_path in library_paths if os.path.exists(library_path) ]))
+
+			subprocess.call([ shutil.which('conda'), 'env', 'config', 'vars', 'set', 'PATH=' + os.pathsep.join(library_paths) ])
+
+	if args.onnxruntime == 'directml':
+		subprocess.call([ shutil.which('pip'), 'install', 'numpy==1.26.4', '--force-reinstall' ])