diffviews / README.md
mckell's picture
Update README.md
99b73e0 verified

A newer version of the Gradio SDK is available: 6.6.0

Upgrade
metadata
title: DiffViews
emoji: 🔬
colorFrom: purple
colorTo: blue
sdk: gradio
sdk_version: 6.3.0
app_file: app.py
python_version: '3.10'
pinned: false
license: mit

DiffViews - Diffusion Activation Visualizer

Interactive visualization of diffusion model activations projected to 2D via UMAP.

Features

  • Explore activation space of diffusion models
  • Select points and find nearest neighbors
  • Generate images from averaged neighbor activations
  • Visualize denoising trajectories

Usage

  1. Hover over points to preview samples
  2. Click to select a point
  3. Click nearby points or use "Suggest KNN" to add neighbors
  4. Click "Generate from Neighbors" to create new images

Note

First launch downloads ~2.5GB of data and checkpoints. Generation on CPU takes ~30-60s per image.