app.py

import gradio as gr
import numpy as np
import random

from diffusers import StableDiffusionPipeline, ControlNetModel, StableDiffusionControlNetPipeline
from peft import PeftModel, PeftConfig
import torch

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

# Model list including your LoRA model
MODEL_LIST = [
    "CompVis/stable-diffusion-v1-4",
    "stabilityai/sdxl-turbo",
    "runwayml/stable-diffusion-v1-5",
    "stabilityai/stable-diffusion-2-1",
    "YaArtemNosenko/dino_stickers",
]

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

# Cache to avoid re-initializing pipelines repeatedly
model_cache = {}

def load_pipeline(model_id,
                  lora_scale,
                  controlnet_checkbox,
                  controlnet_mode,
                  ip_adapter_checkbox,
                  ip_adapter_scale
                  ):
    """
    Loads or retrieves a cached DiffusionPipeline.
    
    If the chosen model is your LoRA adapter, then load the base model 
    (CompVis/stable-diffusion-v1-4) and apply the LoRA weights.
    """
    if model_id in model_cache:
        return model_cache[model_id]


    if controlnet_checkbox:
        if controlnet_mode == "depth_map":
            controlnet = ControlNetModel.from_pretrained(
                "lllyasviel/sd-controlnet-depth",
                cache_dir="./models_cache",
                torch_dtype=torch_dtype
            )
        elif controlnet_mode == "pose_estimation":
            controlnet = ControlNetModel.from_pretrained(
                "lllyasviel/sd-controlnet-openpose",
                cache_dir="./models_cache",
                torch_dtype=torch_dtype
            )
        elif controlnet_mode == "normal_map":
            controlnet = ControlNetModel.from_pretrained(
                "lllyasviel/sd-controlnet-normal",
                cache_dir="./models_cache",
                torch_dtype=torch_dtype
            )
        elif controlnet_mode == "scribbles":
            controlnet = ControlNetModel.from_pretrained(
                "lllyasviel/sd-controlnet-scribble",
                cache_dir="./models_cache",
                torch_dtype=torch_dtype
            )
        else:
            controlnet = ControlNetModel.from_pretrained(
                "lllyasviel/sd-controlnet-canny",
                cache_dir="./models_cache",
                torch_dtype=torch_dtype
            )
        if model_id == "YaArtemNosenko/dino_stickers":
            # Use the specified base model for your LoRA adapter.
            base_model = "CompVis/stable-diffusion-v1-4"
            # Load the LoRA weights
            pipe = StableDiffusionControlNetPipeline.from_pretrained(base_model, 
                                                                     controlnet=controlnet,
                                                                     torch_dtype=torch_dtype, 
                                                                     safety_checker=None).to(device)
            pipe.unet = PeftModel.from_pretrained(
                pipe.unet, 
                model_id, 
                subfolder="unet", 
                torch_dtype=torch_dtype
            )
            pipe.text_encoder = PeftModel.from_pretrained(
                pipe.text_encoder, 
                model_id, 
                subfolder="text_encoder", 
                torch_dtype=torch_dtype
            )
        else:
            pipe = StableDiffusionControlNetPipeline.from_pretrained(model_id,
                                                                     controlnet=controlnet,
                                                                     torch_dtype=torch_dtype,
                                                                     safety_checker=None).to(device)
        # params['image'] = controlnet_image
        # params['controlnet_conditioning_scale'] = float(controlnet_strength)
    else:
        if model_id == "YaArtemNosenko/dino_stickers":
            base_model = "CompVis/stable-diffusion-v1-4"
            pipe = StableDiffusionPipeline.from_pretrained(base_model, torch_dtype=torch_dtype)
            # Load the LoRA weights
            pipe.unet = PeftModel.from_pretrained(
                                                pipe.unet, 
                                                model_id, 
                                                subfolder="unet", 
                                                torch_dtype=torch_dtype
            )

            pipe.text_encoder = PeftModel.from_pretrained(
                pipe.text_encoder, 
                model_id, 
                subfolder="text_encoder", 
                torch_dtype=torch_dtype
            )
        else:
            pipe = StableDiffusionPipeline.from_pretrained(model_id,
                                                           torch_dtype=torch_dtype,
                                                           safety_checker=None).to(device)

            pipe.unet.load_state_dict({k: lora_scale * v for k, v in pipe.unet.state_dict().items()})
            pipe.text_encoder.load_state_dict({k: lora_scale * v for k, v in pipe.text_encoder.state_dict().items()})

    if ip_adapter_checkbox:
        pipe.load_ip_adapter("h94/IP-Adapter",
                             subfolder="models",
                             weight_name="ip-adapter-plus_sd15.bin"
                             )
        pipe.set_ip_adapter_scale(ip_adapter_scale)
        # params['ip_adapter_image'] = ip_adapter_image

    pipe.to(device)
    model_cache[model_id] = pipe
    return pipe

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

def infer(
    model_id,
    prompt,
    negative_prompt,
    seed,
    randomize_seed,
    width,
    height,
    guidance_scale,
    num_inference_steps,
    lora_scale,                         # New parameter for adjusting LoRA scale
    controlnet_checkbox=False,          # используем ли мы controlnet
    controlnet_conditioning_scale=0.0,  # вес для controlnet
    controlnet_mode="edge_detection",   # вариант controlnet
    controlnet_image=None,              # картинка для controlnet
    ip_adapter_checkbox=False,          # используется ли ip адаптера
    ip_adapter_scale=0.0,               # вес для ip адаптера
    ip_adapter_image=None,              # картинка для ip адаптера
    progress=gr.Progress(track_tqdm=True),
):
    # Load the pipeline for the chosen model
    generator = torch.Generator(device=device).manual_seed(seed)
    params = {'prompt': prompt,
              'negative_prompt': negative_prompt,
              'guidance_scale': guidance_scale,
              'num_inference_steps': num_inference_steps,
              'width': width,
              'height': height,
              'generator': generator
              }
    pipe = load_pipeline(model_id,
                         lora_scale,
                         controlnet_checkbox,
                         controlnet_mode,
                         ip_adapter_checkbox,
                         ip_adapter_scale
                         )

    if randomize_seed:
        seed = random.randint(0, MAX_SEED)

    # If using the LoRA model, update the LoRA scale if supported.
    # if model_id == "YaArtemNosenko/dino_stickers":
    #     # This assumes your pipeline's unet has a method to update the LoRA scale.
    #     if hasattr(pipe.unet, "set_lora_scale"):
    #         pipe.unet.set_lora_scale(lora_scale)
    #     else:
    #         print("Warning: LoRA scale adjustment method not found on UNet.")
    # если используем controlnet
    if controlnet_checkbox:
        params['image'] = controlnet_image
        params['controlnet_conditioning_scale'] = float(controlnet_conditioning_scale)
    # если используем IP адаптер
    if ip_adapter_checkbox:
        params['ip_adapter_image'] = ip_adapter_image

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


def controlnet_params(show_extra):
    return gr.update(visible=show_extra)


examples = [
    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
    "An astronaut riding a green horse",
    "A delicious ceviche cheesecake slice",
]

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")

        with gr.Row():
            # Dropdown to select the model from Hugging Face
            model_id = gr.Dropdown(
                label="Model",
                choices=MODEL_LIST,
                value=MODEL_LIST[0],  # Default model
            )

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

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

            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=20.0,
                    step=0.5,
                    value=7.0,
                )

                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=100,
                    step=1,
                    value=20,
                )

            # New slider for LoRA scale.
            lora_scale = gr.Slider(
                label="LoRA Scale",
                minimum=0.0,
                maximum=2.0,
                step=0.1,
                value=1.0,
                info="Adjust the influence of the LoRA weights",
            )
        with gr.Row():
            controlnet_checkbox = gr.Checkbox(
                label="ControlNet",
                value=False
            )
            with gr.Column(visible=False) as controlnet_params:
                controlnet_conditioning_scale = gr.Slider(
                    label="ControlNet conditioning scale",
                    minimum=0.0,
                    maximum=1.0,
                    step=0.01,
                    value=1.0,  
                )
                controlnet_mode = gr.Dropdown(
                    label="ControlNet mode",
                    choices=["edge_detection", 
                             "depth_map",
                             "pose_estimation", 
                             "normal_map",
                             "scribbles"],
                    value="edge_detection",
                    max_choices=1
                )
                controlnet_image = gr.Image(
                    label="ControlNet condition image",
                    type="pil",
                    format="png"
                )
            controlnet_checkbox.change(
                    fn=lambda x: gr.Row.update(visible=x),
                    inputs=controlnet_checkbox,
                    outputs=controlnet_params
            )
        with gr.Row():
            ip_adapter_checkbox = gr.Checkbox(
                label="IPAdapter",
                value=False
            )
            with gr.Column(visible=False) as ip_adapter_params:
                ip_adapter_scale = gr.Slider(
                    label="IPAdapter scale",
                    minimum=0.0,
                    maximum=1.0,
                    step=0.01,
                    value=1.0,  
                )
                ip_adapter_image = gr.Image(
                    label="IPAdapter condition image",
                    type="pil"
                )
            ip_adapter_checkbox.change(
                fn=lambda x: gr.Row.update(visible=x),
                inputs=ip_adapter_checkbox,
                outputs=ip_adapter_params
            )

        gr.Examples(examples=examples, inputs=[prompt])
    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,
            lora_scale,  # Pass the new slider value
            controlnet_checkbox,
            controlnet_conditioning_scale,
            controlnet_mode,
            controlnet_image,
            ip_adapter_checkbox,
            ip_adapter_scale,
            ip_adapter_image
        ],
        outputs=[result, seed],
    )

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