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) 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") generator = torch.Generator(device="cuda") #pipe.enable_model_cpu_offload() def infer(use_custom_model, model_name, image_in, prompt, negative_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, use_auth_token=True) prompt = prompt negative_prompt = negative_prompt if preprocessor == "canny": 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, 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, 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" css=""" #col-container{ margin: 0 auto; max-width: 680px; text-align: left; } """ with gr.Blocks(css=css) as demo: with gr.Column(elem_id="col-container"): gr.Markdown(""" # SD-XL Control LoRas Use StableDiffusion XL with ControlNet pretrained LoRas """) use_custom_model = gr.Checkbox(label="Use a public custom model ?(optional)", value=False, info="To use a private model, you'll prefer to duplicate the space with your own access token.") model_name = gr.Textbox(label="Custom Model to use", placeholder="username/my_custom_public_model") image_in = gr.Image(source="upload", type="filepath") with gr.Row(): with gr.Column(): prompt = gr.Textbox(label="Prompt") negative_prompt = gr.Textbox(label="Negative prompt", value="extra digit, fewer digits, cropped, worst quality, low quality, glitch, deformed, mutated, ugly, disfigured") guidance_scale = gr.Slider(label="Guidance Scale", minimum=1.0, maximum=10.0, step=0.1, value=7.5) with gr.Column(): preprocessor = gr.Dropdown(label="Preprocessor", choices=["canny"], value="canny", interactive=False, info="For the moment, only canny is available") 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, negative_prompt, preprocessor, controlnet_conditioning_scale, guidance_scale, seed], outputs = [result] ) demo.queue().launch()