app.py

#!/usr/bin/env python

from __future__ import annotations

import os
import random

import gradio as gr
import numpy as np
import PIL.Image
import requests
import spaces
import torch
from diffusers import AutoencoderKL, DiffusionPipeline


DESCRIPTION = "# AI 作画"
if not torch.cuda.is_available():
    DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"

MAX_SEED = np.iinfo(np.int32).max
CACHE_EXAMPLES = torch.cuda.is_available() and os.getenv("CACHE_EXAMPLES") == "1"
MAX_IMAGE_SIZE = int(os.getenv("MAX_IMAGE_SIZE", "1024"))
USE_TORCH_COMPILE = os.getenv("USE_TORCH_COMPILE") == "1"
ENABLE_CPU_OFFLOAD = os.getenv("ENABLE_CPU_OFFLOAD") == "1"
ENABLE_REFINER = os.getenv("ENABLE_REFINER", "1") == "1"

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
if torch.cuda.is_available():
    vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
    pipe = DiffusionPipeline.from_pretrained(
        "stabilityai/stable-diffusion-xl-base-1.0",
        vae=vae,
        torch_dtype=torch.float16,
        use_safetensors=True,
        variant="fp16",
    )
    if ENABLE_REFINER:
        refiner = DiffusionPipeline.from_pretrained(
            "stabilityai/stable-diffusion-xl-refiner-1.0",
            vae=vae,
            torch_dtype=torch.float16,
            use_safetensors=True,
            variant="fp16",
        )

    if ENABLE_CPU_OFFLOAD:
        pipe.enable_model_cpu_offload()
        if ENABLE_REFINER:
            refiner.enable_model_cpu_offload()
    else:
        pipe.to(device)
        if ENABLE_REFINER:
            refiner.to(device)

    if USE_TORCH_COMPILE:
        pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
        if ENABLE_REFINER:
            refiner.unet = torch.compile(refiner.unet, mode="reduce-overhead", fullgraph=True)


def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    return seed

def translateEN(zh):
    if zh:
        result = requests.post(
            "https://api-free.deepl.com/v2/translate", 
            params={ 
                "auth_key": "e8b4d428-ada5-3f8d-f965-bad01e8a06c1:fx", 
                "target_lang": "EN-US", 
                "text": zh})
        return result.json()["translations"][0]["text"]

def process_text(prompt):
    if prompt:
        print("中文提示词: \n", prompt)
        prompt_trans = translateEN(prompt)
        print("prompt: \n", prompt_trans)
        return prompt_trans

@spaces.GPU
def generate(
    prompt: str,
    # size_option: str = "竖版",
    negative_prompt: str = "",
    prompt_2: str = "",
    negative_prompt_2: str = "",
    use_negative_prompt: bool = False,
    use_prompt_2: bool = False,
    use_negative_prompt_2: bool = False,
    seed: int = 0,
    width: int = 736,
    height: int = 1024,
    guidance_scale_base: float = 5.0,
    guidance_scale_refiner: float = 5.0,
    num_inference_steps_base: int = 25,
    num_inference_steps_refiner: int = 25,
    apply_refiner: bool = False,
) -> PIL.Image.Image:
    generator = torch.Generator().manual_seed(seed)

    if not use_negative_prompt:
        negative_prompt = None  # type: ignore
    if not use_prompt_2:
        prompt_2 = None  # type: ignore
    if not use_negative_prompt_2:
        negative_prompt_2 = None  # type: ignore

    
    # if size_option == "横版":
    #     width, height = 1024, 736
    # elif size_option == "竖版":
    #     width, height = 736, 1024
    # elif size_option == "方形":
    #     width, height = 736, 736
    # else:
    #     width, height = 736, 1024  # 可以定义一个默认值
        
    # process_text("里面做一个测试")
    # print("prompt是：", prompt)
    # print("negative_prompt是：", negative_prompt)
    # print("prompt_2是：", prompt_2)
    # print("negative_prompt_2是：", negative_prompt_2)
    
    if not apply_refiner:
        return pipe(
            prompt=process_text(prompt),
            negative_prompt=process_text(negative_prompt),
            prompt_2=process_text(prompt_2),
            negative_prompt_2=process_text(negative_prompt_2),
            width=width,
            height=height,
            guidance_scale=guidance_scale_base,
            num_inference_steps=num_inference_steps_base,
            generator=generator,
            output_type="pil",
        ).images[0]
    else:
        latents = pipe(
            prompt=process_text(prompt),
            negative_prompt=process_text(negative_prompt),
            prompt_2=process_text(prompt_2),
            negative_prompt_2=process_text(negative_prompt_2),
            width=width,
            height=height,
            guidance_scale=guidance_scale_base,
            num_inference_steps=num_inference_steps_base,
            generator=generator,
            output_type="latent",
        ).images
        image = refiner(
            prompt=process_text(prompt),
            negative_prompt=process_text(negative_prompt),
            prompt_2=process_text(prompt_2),
            negative_prompt_2=process_text(negative_prompt_2),
            guidance_scale=guidance_scale_refiner,
            num_inference_steps=num_inference_steps_refiner,
            image=latents,
            generator=generator,
        ).images[0]
        return image


examples = [
    "宇航员在丛林中，冷色调，柔和的色彩，细节，8k",
    "一只熊猫戴着草帽，在湖面上划船，电影风格，4K",
]

with gr.Blocks(css="style.css") as demo:
    gr.Markdown(DESCRIPTION)
    gr.DuplicateButton(
        value="Duplicate Space for private use",
        elem_id="duplicate-button",
        visible=os.getenv("SHOW_DUPLICATE_BUTTON") == "1",
    )
    with gr.Group():
        with gr.Row():
            prompt = gr.Text(
                label="提示词",
                show_label=False,
                max_lines=1,
                placeholder="输入要生成的画面内容",
                container=False,
            )
            run_button = gr.Button("生成", scale=0)
        result = gr.Image(label="生成结果", show_label=False)

        # # 使用 Radio 组件替代两个 Slider 组件
        # size_option = gr.Radio(choices=["横版", "竖版", "方形"], label="选择尺寸", value="竖版")
        
    with gr.Accordion("高级选项", open=False):
        with gr.Row():
            use_negative_prompt = gr.Checkbox(label="使用反向提示词", value=False)
            use_prompt_2 = gr.Checkbox(label="使用提示词 2", value=False)
            use_negative_prompt_2 = gr.Checkbox(label="使用反向提示词 2", value=False)
        negative_prompt = gr.Text(
            label="反向提示词",
            max_lines=1,
            placeholder="输入不想在画面中出现的内容，比如：“胡子”，“人群”",
            visible=False,
        )
        prompt_2 = gr.Text(
            label="提示词 2",
            max_lines=1,
            placeholder="输入你的提示词",
            visible=False,
        )
        negative_prompt_2 = gr.Text(
            label="反向提示词 2",
            max_lines=1,
            placeholder="输入你的反向提示词",
            visible=False,
        )

        seed = gr.Slider(
            label="种子数",
            minimum=0,
            maximum=MAX_SEED,
            step=1,
            value=0,
        )
        randomize_seed = gr.Checkbox(label="随机种子数", value=True)
        
        with gr.Row():
            width = gr.Slider(
                label="宽度",
                minimum=256,
                maximum=MAX_IMAGE_SIZE,
                step=32,
                value=736,
            )
            height = gr.Slider(
                label="高度",
                minimum=256,
                maximum=MAX_IMAGE_SIZE,
                step=32,
                value=1024,
            )
        
        apply_refiner = gr.Checkbox(label="增加精炼模型（refiner）", value=False, visible=ENABLE_REFINER)
        
        with gr.Row():
            guidance_scale_base = gr.Slider(
                label="提示词相关性",
                minimum=1,
                maximum=20,
                step=0.1,
                value=7.5,
            )
            num_inference_steps_base = gr.Slider(
                label="模型迭代步数",
                minimum=10,
                maximum=100,
                step=1,
                value=25,
            )
        with gr.Row(visible=False) as refiner_params:
            guidance_scale_refiner = gr.Slider(
                label="提示词相关性（refiner）",
                minimum=1,
                maximum=20,
                step=0.1,
                value=7.5,
            )
            num_inference_steps_refiner = gr.Slider(
                label="模型迭代步数（refiner）",
                minimum=10,
                maximum=100,
                step=1,
                value=25,
            )
    
    gr.Examples(
        label="例子",
        examples=examples,
        inputs=prompt,
        outputs=result,
        fn=generate,
        cache_examples=CACHE_EXAMPLES,
    )

    use_negative_prompt.change(
        fn=lambda x: gr.update(visible=x),
        inputs=use_negative_prompt,
        outputs=negative_prompt,
        queue=False,
        api_name=False,
    )
    use_prompt_2.change(
        fn=lambda x: gr.update(visible=x),
        inputs=use_prompt_2,
        outputs=prompt_2,
        queue=False,
        api_name=False,
    )
    use_negative_prompt_2.change(
        fn=lambda x: gr.update(visible=x),
        inputs=use_negative_prompt_2,
        outputs=negative_prompt_2,
        queue=False,
        api_name=False,
    )
    apply_refiner.change(
        fn=lambda x: gr.update(visible=x),
        inputs=apply_refiner,
        outputs=refiner_params,
        queue=False,
        api_name=False,
    )

    gr.on(
        triggers=[
            prompt.submit,
            negative_prompt.submit,
            prompt_2.submit,
            negative_prompt_2.submit,
            run_button.click,
        ],
        fn=randomize_seed_fn,
        inputs=[seed, randomize_seed],
        outputs=seed,
        queue=False,
        api_name=False,
    ).then(
        fn=generate,
        inputs=[
            prompt,
            # size_option,
            negative_prompt,
            prompt_2,
            negative_prompt_2,
            use_negative_prompt,
            use_prompt_2,
            use_negative_prompt_2,
            seed,
            width,
            height,
            guidance_scale_base,
            guidance_scale_refiner,
            num_inference_steps_base,
            num_inference_steps_refiner,
            apply_refiner,
        ],
        outputs=result,
        api_name="run",
    )

if __name__ == "__main__":
    demo.queue(max_size=30).launch(max_threads=2)