app.py

import gradio as gr
import numpy as np
import random

import spaces #[uncomment to use ZeroGPU]
from diffusers import DiffusionPipeline
import torch

from peft import PeftModel, LoraConfig
import os

def get_lora_sd_pipeline(
    ckpt_dir='./lora', 
    base_model_name_or_path=None, 
    dtype=torch.float16, 
    adapter_name="default"
    ):

    unet_sub_dir = os.path.join(ckpt_dir, "unet")
    text_encoder_sub_dir = os.path.join(ckpt_dir, "text_encoder")
    
    if os.path.exists(text_encoder_sub_dir) and base_model_name_or_path is None:
        config = LoraConfig.from_pretrained(text_encoder_sub_dir)
        base_model_name_or_path = config.base_model_name_or_path
    
    if base_model_name_or_path is None:
        raise ValueError("Please specify the base model name or path")
    
    pipe = DiffusionPipeline.from_pretrained(base_model_name_or_path, torch_dtype=dtype)
    before_params = pipe.unet.parameters()
    pipe.unet = PeftModel.from_pretrained(pipe.unet, unet_sub_dir, adapter_name=adapter_name)
    pipe.unet.set_adapter(adapter_name)
    after_params = pipe.unet.parameters()
    print("Parameters changed:", any(torch.any(b != a) for b, a in zip(before_params, after_params)))
    
    if os.path.exists(text_encoder_sub_dir):
        pipe.text_encoder = PeftModel.from_pretrained(pipe.text_encoder, text_encoder_sub_dir, adapter_name=adapter_name)
    
    if dtype in (torch.float16, torch.bfloat16):
        pipe.unet.half()
        pipe.text_encoder.half()
    
    return pipe

def process_prompt(prompt, tokenizer, text_encoder, max_length=77):
    tokens = tokenizer(prompt, truncation=False, return_tensors="pt")["input_ids"]
    chunks = [tokens[:, i:i + max_length] for i in range(0, tokens.shape[1], max_length)]
    
    with torch.no_grad():
        embeds = [text_encoder(chunk.to(text_encoder.device))[0] for chunk in chunks]
    
    return torch.cat(embeds, dim=1)

def align_embeddings(prompt_embeds, negative_prompt_embeds):
    max_length = max(prompt_embeds.shape[1], negative_prompt_embeds.shape[1])
    return torch.nn.functional.pad(prompt_embeds, (0, 0, 0, max_length - prompt_embeds.shape[1])), \
           torch.nn.functional.pad(negative_prompt_embeds, (0, 0, 0, max_length - negative_prompt_embeds.shape[1]))

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

model_id_default = "sd-legacy/stable-diffusion-v1-5"
model_dropdown = ['stabilityai/sdxl-turbo', 'CompVis/stable-diffusion-v1-4', 'sd-legacy/stable-diffusion-v1-5']

model_lora_default = "lora"

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


@spaces.GPU #[uncomment to use ZeroGPU]
def infer(
    prompt,
    negative_prompt,
    randomize_seed,
    width=512,
    height=512,
    model_repo_id=model_id_default,
    seed=42,
    guidance_scale=7,
    num_inference_steps=20,
    model_lora_id=model_lora_default,
    lora_scale=0.5,
    progress=gr.Progress(track_tqdm=True),
    ):
        
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)

    generator = torch.Generator().manual_seed(seed)
    
    # добавляем обновление pipe по условию
    if model_repo_id != model_id_default:
        pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype).to(device)
        prompt_embeds = process_prompt(prompt, pipe.tokenizer, pipe.text_encoder)
        negative_prompt_embeds = process_prompt(negative_prompt, pipe.tokenizer, pipe.text_encoder)
        prompt_embeds, negative_prompt_embeds = align_embeddings(prompt_embeds, negative_prompt_embeds)
    else:
        # добавляем lora
        pipe = get_lora_sd_pipeline(ckpt_dir='./' + model_lora_id, base_model_name_or_path=model_id_default, dtype=torch_dtype).to(device)
        prompt_embeds = process_prompt(prompt, pipe.tokenizer, pipe.text_encoder)
        negative_prompt_embeds = process_prompt(negative_prompt, pipe.tokenizer, pipe.text_encoder)
        prompt_embeds, negative_prompt_embeds = align_embeddings(prompt_embeds, negative_prompt_embeds)
        print(f"LoRA adapter loaded: {pipe.unet.active_adapters}")
        print(f"LoRA scale applied: {lora_scale}")
        pipe.fuse_lora(lora_scale=lora_scale)

    # на вызов pipe с эмбеддингами
    params = {
        'prompt_embeds': prompt_embeds,
        'negative_prompt_embeds': negative_prompt_embeds,
        'guidance_scale': guidance_scale,
        'num_inference_steps': num_inference_steps,
        'width': width,
        'height': height,
        'generator': generator,
    }

    return pipe(**params).images[0], seed    


examples = [
    "A cartoon-style sticker of Elon Musk shaking hands with Donald Trump. Both figures have exaggerated facial expressions, with Musk grinning confidently and Trump giving a signature thumbs-up. The background features a patriotic red, white, and blue color scheme with fireworks exploding behind them.",
    "A cyberpunk-themed cartoon sticker of Elon Musk standing atop a futuristic Tesla spaceship. He wears a sleek, neon-lit jacket with glowing circuits, while the city skyline behind him is filled with holographic billboards displaying SpaceX and Neuralink logos. His sunglasses reflect the distant stars, adding to the sci-fi aesthetic.",
    "A medieval fantasy sticker of Elon Musk depicted as a wizard. He holds a glowing blue orb in one hand and a spellbook in the other, wearing a long, starry robe with intricate golden details. His expression is both wise and mischievous, as if he's about to reveal the secrets of the universe. The background features a mystical castle and a dragon flying in the sky.",
    "A sticker of Elon Musk dressed as a cowboy in the Wild West. He wears a wide-brimmed hat, leather boots, and a long trench coat, standing in front of a saloon with a SpaceX rocket docked nearby instead of a horse. A wanted poster on the wall reads 'Wanted: Mars Pioneer', adding to the playful western theme.",
    "A parody cartoon sticker of Elon Musk arm-wrestling a robotic version of himself. The robot Musk has glowing red eyes and mechanical arms, while the real Musk smirks confidently. Sparks fly from the table as the intense match unfolds, and the background features a neon sign that reads 'Tesla vs. AI: Ultimate Showdown'."
]

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("# Generate LoRa stickers")

        with gr.Row():
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt",
                container=False,
            )

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

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

        with gr.Accordion("Advanced Settings", open=False):
            
            model_repo_id = gr.Dropdown(
                label="Model Id",
                choices=model_dropdown,
                info="Choose model",
                visible=True,
                allow_custom_value=True,
                value=model_id_default,
            )  

            negative_prompt = gr.Text(
                label="Negative prompt",
                max_lines=1,
                placeholder="Enter a negative prompt",
                visible=True,
                value="blurry, low quality, distorted, low resolution, medical mask"
            )

            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=42,
            )

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

            with gr.Row():   
                width = gr.Slider(
                    label="Width",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=512,  # Replace with defaults that work for your model
                )

                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=512,  # Replace with defaults that work for your model
                )

        gr.Examples(examples=examples, inputs=[prompt])
    gr.on(
        triggers=[run_button.click, prompt.submit],
        fn=infer,
        inputs=[
            prompt,
            negative_prompt,
            randomize_seed,
            width,
            height,
            model_repo_id,
            seed
        ],
        outputs=[result, seed],
    )

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