import gradio as gr import numpy as np import random from typing import Optional # import spaces #[uncomment to use ZeroGPU] from diffusers import StableDiffusionPipeline, StableDiffusionControlNetPipeline from diffusers import ControlNetModel from peft import PeftModel, LoraConfig from rembg import new_session, remove from PIL import Image as PILImage import cv2 import torch device = "cuda" if torch.cuda.is_available() else "cpu" if torch.cuda.is_available(): torch_dtype = torch.float16 else: torch_dtype = torch.float32 import os # import torch MAX_SEED = np.iinfo(np.int32).max MAX_IMAGE_SIZE = 1024 CONTROL_MODE_MODEL = { "Canny Ege Detection" : "lllyasviel/control_v11p_sd15_canny", "Pixel to Pixel": "lllyasviel/control_v11e_sd15_ip2p", "M-LSD Line detection" : "lllyasviel/control_v11p_sd15_mlsd", "HED edge detection (soft edge)" : "lllyasviel/control_sd15_hed", "Midas depth estimationn" : "lllyasviel/control_v11f1p_sd15_depth", "Surface Normal Estimation" : "lllyasviel/control_v11p_sd15_normalbae", "Scribble-Based Generation" : "lllyasviel/control_v11p_sd15_scribble", "Semantic segmentation" : "lllyasviel/control_v11p_sd15_seg", "OpenPose pose detection" : "lllyasviel/control_v11p_sd15_openpose", "Line Art Generation": "lllyasviel/control_v11p_sd15_lineart", } # @spaces.GPU #[uncomment to use ZeroGPU] def infer( prompt: str, negative_prompt : str, width, height, lscale=0.0, remove_background=False, controlnet_enabled=False, controlnet_strength=0.0, controlnet_mode=None, controlnet_image=None, ip_adapter_enabled=False, ip_adapter_scale=0.0, ip_adapter_image=None, model_id: Optional[str] = "CompVis/stable-diffusion-v1-4", seed: Optional[int] = 42, guidance_scale : Optional[int] = 7, num_inference_steps : Optional[int] = 20, progress=gr.Progress(track_tqdm=True), ): generator = torch.Generator().manual_seed(seed) if controlnet_enabled: if not controlnet_image : raise ValueError("controlnet_enabled set to True, but controlnet_image not given") else: controlnet_model = ControlNetModel.from_pretrained(CONTROL_MODE_MODEL.get(controlnet_mode),torch_dtype=torch_dtype) if model_id == "SD-v1-5 + Lora" : pipe=StableDiffusionControlNetPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5",controlnet=controlnet_model, torch_dtype=torch_dtype) pipe.unet = PeftModel.from_pretrained(pipe.unet , "Emilichcka/diffusion_lora_funny_cat", subfolder="unet", torch_dtype=torch_dtype) pipe.text_encoder = PeftModel.from_pretrained(pipe.text_encoder,"Emilichcka/diffusion_lora_funny_cat", subfolder="text_encoder", torch_dtype=torch_dtype) else: pipe=StableDiffusionControlNetPipeline.from_pretrained(model_id, controlnet=controlnet_model, torch_dtype=torch_dtype) else: if model_id == "SD-v1-5 + Lora" : pipe=StableDiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5",torch_dtype=torch_dtype) pipe.unet = PeftModel.from_pretrained(pipe.unet , "Emilichcka/diffusion_lora_funny_cat", subfolder="unet", torch_dtype=torch_dtype) pipe.text_encoder = PeftModel.from_pretrained(pipe.text_encoder,"Emilichcka/diffusion_lora_funny_cat", subfolder="text_encoder", torch_dtype=torch_dtype) else: pipe=StableDiffusionPipeline.from_pretrained(model_id) if ip_adapter_enabled: ip_adapter_scale = float(ip_adapter_scale) pipe.load_ip_adapter("h94/IP-Adapter",subfolder="models", weight_name="ip-adapter-plus_sd15.bin", torch_dtype=torch_dtype) pipe.set_ip_adapter_scale(ip_adapter_scale) if controlnet_image!= None: controlnet_image = np.array(controlnet_image) low_threshold = 100 high_threshold = 200 controlnet_image = cv2.Canny(controlnet_image, low_threshold, high_threshold) controlnet_image = controlnet_image[:, :, None] controlnet_image = np.concatenate([controlnet_image, controlnet_image, controlnet_image], axis=2) controlnet_image = PILImage.fromarray(controlnet_image) pipe = pipe.to(device) image = pipe( prompt=prompt, image=controlnet_image, negative_prompt=negative_prompt, guidance_scale=guidance_scale, num_inference_steps=num_inference_steps, width=width, height=height, generator=generator, cross_attention_kwargs={"scale": lscale}, controlnet_conditioning_scale=controlnet_strength, ip_adapter_image=ip_adapter_image, ).images[0] if remove_background: image = remove(image) return image, seed 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: 880px; } """ default_model_id_choice = [ "stable-diffusion-v1-5/stable-diffusion-v1-5", "CompVis/stable-diffusion-v1-4", "SD-v1-5 + Lora", "nota-ai/bk-sdm-small", ] def update_controlnet_visibility(controlnet_enabled): return gr.update(visible=controlnet_enabled), gr.update(visible=controlnet_enabled), gr.update(visible=controlnet_enabled) def update_ip_adapter_visibility(ip_adapter_enabled): return gr.update(visible=ip_adapter_enabled), gr.update(visible=ip_adapter_enabled) with gr.Blocks(css=css) as demo: with gr.Column(elem_id="col-container"): gr.Markdown(" # Text-to-Image Gradio Template") with gr.Row(): model_id = gr.Dropdown( label="Model Selection", choices=default_model_id_choice, value="SD-v1-5 + Lora", ) seed = gr.Slider( label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=42, ) 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.Row(): remove_background = gr.Checkbox(label="Remove Background", value=False) controlnet_enabled = gr.Checkbox(label="Enable ControlNet", value=False) ip_adapter_enabled = gr.Checkbox(label="Enable IP-Adapter", value=False) with gr.Accordion("ControlNet Settings", open=False): gr.Markdown("Enable ControlNet to use settings", visible=True) with gr.Row(): controlNet_strength = gr.Slider( label="ControlNet scale", minimum=0.0, maximum=1.0, step=0.05, value=0.75, visible=False, interactive=True, ) controlNet_mode = gr.Dropdown( label="ControlNet Mode", choices=list(CONTROL_MODE_MODEL.keys()), visible=False, interactive=True, ) controlNet_image = gr.Image(label="ControlNet Image", type="pil", interactive=True, visible=False) with gr.Accordion("IP-Adapter Settings", open=False): gr.Markdown("Enable IP-Adapter to use settings", visible=True) with gr.Row(): ip_adapter_scale = gr.Slider( label="IP-Adapter Scale", minimum=0.0, maximum=2.0, step=0.05, value=1.0, visible=False, interactive=True, ) ip_adapter_image = gr.Image(label="IP-Adapter Image", type="pil",interactive=True, visible=False) with gr.Accordion("Advanced Settings", open=False): negative_prompt = gr.Text( label="Negative prompt", max_lines=1, placeholder="Enter a negative prompt", ) 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", interactive=True, ) 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 ) with gr.Row(): guidance_scale = gr.Slider( label="Guidance scale", minimum=0.0, maximum=10.0, step=0.1, value=10.0, # Replace with defaults that work for your model ) num_inference_steps = gr.Slider( label="Number of inference steps", minimum=1, maximum=50, step=1, value=30, # 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, width, height, lora_scale, remove_background, controlnet_enabled, controlNet_strength, controlNet_mode, controlNet_image, ip_adapter_enabled, ip_adapter_scale, ip_adapter_image, model_id, seed, guidance_scale, num_inference_steps, ], outputs=[result, seed], ) controlnet_enabled.change( fn=update_controlnet_visibility, inputs=[controlnet_enabled], outputs=[controlNet_strength, controlNet_mode, controlNet_image], ) ip_adapter_enabled.change( fn=update_ip_adapter_visibility, inputs=[ip_adapter_enabled], outputs=[ip_adapter_scale, ip_adapter_image], ) if __name__ == "__main__": demo.launch(share=True)