app.py

import gradio as gr
import numpy as np
import random
from rembg import remove
from copy import deepcopy

# import spaces  # [uncomment to use ZeroGPU]
from diffusers import StableDiffusionPipeline
import torch
from peft import PeftModel, PeftConfig

import os
os.environ['TRANSFORMERS_OFFLINE'] = '0'
os.environ['HF_HUB_OFFLINE'] = '0'

device = "cuda" if torch.cuda.is_available() else "cpu"

MODEL_OPTIONS = [
    "Stable Diffusion v1-4",
    "Chris the mouse Adapter" 
]

DEFAULT_MODEL_ID = "Stable Diffusion v1-4"

if torch.cuda.is_available():
    torch_dtype = torch.float16
else:
    torch_dtype = torch.float32

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1024

PIPELINES = {} 

def load_pipelines():
    # SD v1-4
    mid = "CompVis/stable-diffusion-v1-4"
    pipe = StableDiffusionPipeline.from_pretrained(
        mid, 
        torch_dtype=torch_dtype,
        use_safetensors=True
    )
    pipe = pipe.to(device)
    PIPELINES["Stable Diffusion v1-4"] = pipe
    
    # lora adapter
    lora_adapter_id = "iresidentevil/lora-sd-v1-4-chris-the-mouse"
    pipe_lora = deepcopy(pipe)
    pipe_lora.unet = PeftModel.from_pretrained(
            pipe_lora.unet, 
            lora_adapter_id, 
            adapter_name="default",
            subfolder="unet"
        )
    pipe_lora.text_encoder = PeftModel.from_pretrained(
        pipe_lora.text_encoder, 
        lora_adapter_id,
        adapter_name="default", 
        subfolder="text_encoder"
    )
    pipe_lora = pipe_lora.to(device)
    PIPELINES["Chris the mouse Adapter"] = pipe_lora
    

# Вызываем сразу при импорте (на сборке образа и при старте Space)
load_pipelines()

def remove_background(image):
    """Удаление фона"""
    if image is None:
        return None
    result = remove(image)
    return result

# @spaces.GPU  # [uncomment to use ZeroGPU]
def infer(
    model_id, 
    prompt,
    negative_prompt,
    seed,
    randomize_seed,
    width,
    height,
    guidance_scale,
    num_inference_steps,
    remove_bg,  # параметр для удаления фона
    lora_scale,  # параметр для масштаба LoRA
    progress=gr.Progress(track_tqdm=True),
):
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)

    generator = torch.Generator().manual_seed(seed)
    
    pipe = PIPELINES[model_id]
    
    # Применяем масштаб LoRA если выбран адаптер и scale != 1.0
    if "Adapter" in model_id and lora_scale != 1.0:
        pipe.unet.scale_layer(lora_scale)
        pipe.text_encoder.scale_layer(lora_scale)

    image = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        width=width,
        height=height,
        generator=generator,
    ).images[0]

    # Удаляем фон если выбрана соответствующая опция
    if remove_bg:
        image = remove_background(image)

    return image, seed

examples = [
    "chris_the_mouse with wide, surprised eyes, single black nostril, shaded gray body, and exaggeratedly large pink ears on a plain black background, drawn in a minimalist cartoon style.",
    "chris_the_mouse with one eye wide open, mouth agape in shock, red tongue showing, left ear bent, bold black outline, simple cartoon style, white background.",
    "chris_the_mouse with large pink ears, wide black eyes gazing upward, clasped hands, and a small smile, set against a plain background in a cartoon sticker style.",
]

css = """
#col-container {
    margin: 0 auto;
    max-width: 640px;
}
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.Markdown(" # Text-to-Image Gradio Template")
        
        model_id = gr.Dropdown(
            choices=MODEL_OPTIONS,
            label="Model",
            value=DEFAULT_MODEL_ID,
        )
        
        with gr.Row():
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt",
                container=False,
            )

            negative_prompt = gr.Text(
                label="Negative prompt",
                max_lines=1,
                placeholder="Enter a negative prompt",
                visible=True,
            )

            remove_bg = gr.Checkbox(
                    label="Delete background?",
                    value=False,
                    info="Remove background using rembg"
                )

            lora_scale = gr.Slider(
                    label="LoRA scale",
                    minimum=0.0,
                    maximum=2.0,
                    step=0.1,
                    value=1.0,
                    visible=False,  
                    info="Adjust LoRA adapter strength"
                )

            run_button = gr.Button("Run", scale=0, variant="primary")

        result = gr.Image(label="Result", show_label=False)

        with gr.Accordion("Advanced Settings", open=False):
            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=0,
            )

            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)

            with gr.Row():
                width = gr.Slider(
                    label="Width",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )

                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )

            with gr.Row():
                guidance_scale = gr.Slider(
                    label="Guidance scale",
                    minimum=0.0,
                    maximum=10.0,
                    step=0.1,
                    value=7.5,  
                )

                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=20,  
                )

        gr.Examples(examples=examples, inputs=[prompt])
    
    # Функция для показа/скрытия слайдера LoRA scale
    def toggle_lora_scale_visibility(model_id):
        if "Adapter" in model_id:
            return gr.Slider(visible=True)
        else:
            return gr.Slider(visible=False)
    
    # Обработчик изменения модели
    model_id.change(
        fn=toggle_lora_scale_visibility,
        inputs=[model_id],
        outputs=[lora_scale]
    )
    
    gr.on(
        triggers=[run_button.click, prompt.submit],
        fn=infer,
        inputs=[
            model_id,
            prompt,
            negative_prompt,
            seed,
            randomize_seed,
            width,
            height,
            guidance_scale,
            num_inference_steps,
            remove_bg,
            lora_scale,
        ],
        outputs=[result, seed],
    )

if __name__ == "__main__":
    demo.launch()