app.py

import gradio as gr
from huggingface_hub import login
import os

hf_token = os.environ.get("HF_TOKEN")
login(token=hf_token)

from diffusers import ControlNetModel, StableDiffusionXLControlNetPipeline, AutoencoderKL
from diffusers.utils import load_image
from PIL import Image
import torch
import numpy as np
import cv2

#vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)

generator = torch.Generator(device="cuda")

#pipe.enable_model_cpu_offload()

def infer(use_custom_model, model_name, image_in, prompt, preprocessor, controlnet_conditioning_scale, guidance_scale, seed):
    if use_custom_model:
        custom_model = model_name

        # This is where you load your trained weights
        pipe.load_lora_weights(custom_model, weight_name="pytorch_lora_weights.safetensors", use_auth_token=True)
    
    prompt = prompt
    negative_prompt = "extra digit, fewer digits, cropped, worst quality, low quality, glitch, deformed, mutated, ugly, disfigured"

    if preprocessor == "canny":

        controlnet = ControlNetModel.from_pretrained(
            "diffusers/controlnet-canny-sdxl-1.0",
            torch_dtype=torch.float16
        )

        pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
            "stabilityai/stable-diffusion-xl-base-1.0",
            controlnet=controlnet,
            #vae=vae,
            torch_dtype=torch.float16, 
            variant="fp16",
            use_safetensors=True
        )
        pipe.to("cuda")
        
        image = load_image(image_in)

        image = np.array(image)
        image = cv2.Canny(image, 100, 200)
        image = image[:, :, None]
        image = np.concatenate([image, image, image], axis=2)
        image = Image.fromarray(image)

    if use_custom_model:
        lora_scale= 0.9

        images = pipe(
            prompt, 
            negative_prompt=negative_prompt, 
            image=image, 
            preprocessor=preprocessor,
            controlnet_conditioning_scale=controlnet_conditioning_scale,
            guidance_scale = guidance_scale,
            num_inference_steps=50,
            generator=generator.manual_seed(seed),
            cross_attention_kwargs={"scale": lora_scale}
        ).images
    else:
        images = pipe(
            prompt, 
            negative_prompt=negative_prompt, 
            image=image, 
            preprocessor=preprocessor,
            controlnet_conditioning_scale=controlnet_conditioning_scale,
            guidance_scale = guidance_scale,
            num_inference_steps=50,
            generator=generator.manual_seed(seed),
        ).images

    images[0].save(f"result.png")

    return f"result.png"

with gr.Blocks() as demo:
    with gr.Column():
        use_custom_model = gr.Checkbox(label="Use a custom model ?", value=False)
        model_name = gr.Textbox(label="Model to use", placeholder="username/my_model")
        image_in = gr.Image(source="upload", type="filepath")
        prompt = gr.Textbox(label="Prompt"),
        preprocessor = gr.Dropdown(label="Preprocessor", choices=["canny"], value="canny")
        guidance_scale = gr.Slider(label="Guidance Scale", minimum=1.0, maximum=10.0, step=0.1, value=7.5, type="float")
        controlnet_conditioning_scale = gr.Slider(label="Controlnet conditioning Scale", minimum=0.1, maximum=0.9, step=0.01, value=0.5, type="float")
        seed = gr.Slider(label="seed", minimum=0, maximum=500000, step=1, value=42)

        submit_btn = gr.Button("Submit")
        result = gr.Image(label="Result")

    submit_btn.click(
        fn = infer,
        inputs = [use_custom_model, model_name, image_in, prompt, preprocessor, controlnet_conditioning_scale, guidance_scale, seed],
        outputs = [result]
    )

demo.queue().launch()