import gradio as gr import numpy as np import random import spaces from diffusers import StableDiffusionXLPipeline, AutoencoderKL, ControlNetModel from diffusers.utils import load_image import torch from typing import Tuple from PIL import Image from controlnet_aux import OpenposeDetector import insightface import onnxruntime ip_adapter_loaded = False device = "cuda" if torch.cuda.is_available() else "cpu" model_repo_id = "RunDiffusion/Juggernaut-XL-v9" # Replace to the model you would like to use vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16) if torch.cuda.is_available(): torch_dtype = torch.float16 else: torch_dtype = torch.float32 pipe = StableDiffusionXLPipeline.from_pretrained( "RunDiffusion/Juggernaut-XL-v9", vae=vae, torch_dtype=torch.float16, custom_pipeline="lpw_stable_diffusion_xl", use_safetensors=True, add_watermarker=False, variant="fp16", ) pipe.to(device) controlnet_openpose = ControlNetModel.from_pretrained( "lllyasviel/control_v11p_sd15_openpose", torch_dtype=torch.float16 ).to(device) openpose_detector = OpenposeDetector.from_pretrained("lllyasviel/ControlNet").to(device) try: pipe.load_ip_adapter("h94/IP-Adapter-FaceID", subfolder="", weight_name="ip-adapter-faceid_sdxl_lora.safetensors") ip_adapter_loaded = True except Exception as e: print(f"Could not load IP-Adapter FaceID. Make sure the model exists and paths are correct: {e}") print("Trying a common alternative: ip-adapter-plus-face_sdxl_vit-h.safetensors") try: pipe.load_ip_adapter("h94/IP-Adapter", subfolder="models", weight_name="ip-adapter-plus-face_sdxl_vit-h.safetensors") except Exception as e2: print(f"Could not load second IP-Adapter variant: {e2}") print("IP-Adapter will not be available. Please check your IP-Adapter setup.") pipe.unload_ip_adapter() MAX_SEED = np.iinfo(np.int32).max MAX_IMAGE_SIZE = 4096 style_list = [ { "name": "(No style)", "prompt": "{prompt}", "negative_prompt": "", }, { "name": "Cinematic", "prompt": "cinematic still {prompt} . emotional, harmonious, vignette, highly detailed, high budget, bokeh, cinemascope, moody, epic, gorgeous, film grain, grainy", "negative_prompt": "anime, cartoon, graphic, text, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured", }, { "name": "Photographic", "prompt": "cinematic photo {prompt} . 35mm photograph, film, bokeh, professional, 4k, highly detailed", "negative_prompt": "drawing, painting, crayon, sketch, graphite, impressionist, noisy, blurry, soft, deformed, ugly", }, { "name": "Anime", "prompt": "anime artwork {prompt} . anime style, key visual, vibrant, studio anime, highly detailed", "negative_prompt": "photo, deformed, black and white, realism, disfigured, low contrast", }, { "name": "Manga", "prompt": "manga style {prompt} . vibrant, high-energy, detailed, iconic, Japanese comic style", "negative_prompt": "ugly, deformed, noisy, blurry, low contrast, realism, photorealistic, Western comic style", }, { "name": "Digital Art", "prompt": "concept art {prompt} . digital artwork, illustrative, painterly, matte painting, highly detailed", "negative_prompt": "photo, photorealistic, realism, ugly", }, { "name": "Pixel art", "prompt": "pixel-art {prompt} . low-res, blocky, pixel art style, 8-bit graphics", "negative_prompt": "sloppy, messy, blurry, noisy, highly detailed, ultra textured, photo, realistic", }, { "name": "Fantasy art", "prompt": "ethereal fantasy concept art of {prompt} . magnificent, celestial, ethereal, painterly, epic, majestic, magical, fantasy art, cover art, dreamy", "negative_prompt": "photographic, realistic, realism, 35mm film, dslr, cropped, frame, text, deformed, glitch, noise, noisy, off-center, deformed, cross-eyed, closed eyes, bad anatomy, ugly, disfigured, sloppy, duplicate, mutated, black and white", }, { "name": "Neonpunk", "prompt": "neonpunk style {prompt} . cyberpunk, vaporwave, neon, vibes, vibrant, stunningly beautiful, crisp, detailed, sleek, ultramodern, magenta highlights, dark purple shadows, high contrast, cinematic, ultra detailed, intricate, professional", "negative_prompt": "painting, drawing, illustration, glitch, deformed, mutated, cross-eyed, ugly, disfigured", }, { "name": "3D Model", "prompt": "professional 3d model {prompt} . octane render, highly detailed, volumetric, dramatic lighting", "negative_prompt": "ugly, deformed, noisy, low poly, blurry, painting", }, ] styles = {k["name"]: (k["prompt"], k["negative_prompt"]) for k in style_list} STYLE_NAMES = list(styles.keys()) DEFAULT_STYLE_NAME = "(No style)" def apply_style(style_name: str, positive: str, negative: str = "") -> Tuple[str, str]: p, n = styles.get(style_name, styles[DEFAULT_STYLE_NAME]) if not negative: negative = "" return p.replace("{prompt}", positive), n + negative @spaces.GPU def infer( prompt, negative_prompt, style, input_image_pose, pose_strength, input_image_face, face_fidelity, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, progress=gr.Progress(track_tqdm=True), ): if randomize_seed: seed = random.randint(0, MAX_SEED) prompt, negative_prompt = apply_style(style, prompt, negative_prompt) generator = torch.Generator().manual_seed(seed) controlnet_images = [] controlnet_conditioning_scales = [] controlnet_models_to_use = [] # Process Pose Reference if input_image_pose: processed_pose_image = openpose_detector(input_image_pose) controlnet_images.append(processed_pose_image) controlnet_conditioning_scales.append(pose_strength) controlnet_models_to_use.append(controlnet_openpose) # Process Face Reference (IP-Adapter) # CORRECTED LINE HERE: Use the 'ip_adapter_loaded' flag if input_image_face and ip_adapter_loaded: # Use the flag to check if IP-Adapter loaded successfully pipe.set_ip_adapter_scale(face_fidelity) else: # If no face input or IP-Adapter failed to load, ensure scale is reset or not applied # This check is for the general 'lora_scale' attribute which IP-Adapter uses if hasattr(pipe, 'lora_scale') and pipe.lora_scale is not None: pipe.set_ip_adapter_scale(0.0) # Reset scale to 0.0 image = pipe( prompt=prompt, negative_prompt=negative_prompt, image=controlnet_images if controlnet_images else None, controlnet_conditioning_scale=controlnet_conditioning_scales if controlnet_conditioning_scales else None, controlnet=controlnet_models_to_use if controlnet_models_to_use else None, # Only pass ip_adapter_image if there's an input_image_face AND the IP-Adapter was successfully loaded ip_adapter_image=input_image_face if input_image_face and ip_adapter_loaded else None, guidance_scale=guidance_scale, num_inference_steps=num_inference_steps, width=width, height=height, generator=generator, ).images[0] return image, seed examples = [ "A stunning woman standing on a beach at sunset, dramatic lighting, highly detailed", "A man in a futuristic city, cyberpunk style, neon lights", "An AI model posing with a friendly robot in a studio, professional photoshoot", ] 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(" # AI Instagram Model Creator") with gr.Row(): prompt = gr.Text( label="Prompt", show_label=False, max_lines=1, placeholder="Describe your AI model and scene (e.g., 'A confident woman in a red dress, city background')", container=False, ) run_button = gr.Button("Generate", scale=0, variant="primary") result = gr.Image(label="Result", show_label=False) with gr.Accordion("Reference Images", open=True): gr.Markdown("Upload images to control pose and face consistency.") input_image_pose = gr.Image(label="Human Pose Reference (for body posture)", type="pil", show_label=True) pose_strength = gr.Slider( label="Pose Control Strength (0.0 = ignore, 1.0 = strict adherence)", minimum=0.0, maximum=1.0, step=0.01, value=0.8, # Good starting point for strong pose control ) gr.Markdown("---") # Separator input_image_face = gr.Image(label="Face Reference (for facial consistency)", type="pil", show_label=True) face_fidelity = gr.Slider( label="Face Fidelity (0.0 = ignore, 1.0 = highly similar)", minimum=0.0, maximum=1.0, step=0.01, value=0.7, # Good starting point for face transfer ) with gr.Row(visible=True): style_selection = gr.Radio( show_label=True, container=True, interactive=True, choices=STYLE_NAMES, value=DEFAULT_STYLE_NAME, label="Image Style", ) with gr.Accordion("Advanced Settings", open=False): negative_prompt = gr.Text( label="Negative prompt", max_lines=1, placeholder="What you DON'T want in the image (e.g., 'deformed, blurry, text')", visible=False, ) seed = gr.Slider( label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0, ) 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=768, ) with gr.Row(): guidance_scale = gr.Slider( label="Guidance scale", minimum=0.0, maximum=20.0, # Increased max for more control step=0.1, value=7.0, ) num_inference_steps = gr.Slider( label="Number of inference steps", minimum=1, maximum=1000, # More typical steps for SDXL (20-50 usually sufficient) step=1, value=60, ) gr.Examples(examples=examples, inputs=[prompt]) gr.on( triggers=[run_button.click, prompt.submit], fn=infer, inputs=[ prompt, negative_prompt, style_selection, input_image_pose, pose_strength, input_image_face, face_fidelity, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, ], outputs=[result, seed], ) if __name__ == "__main__": demo.launch(share=True)