import gradio as gr import numpy as np import random # import spaces #[uncomment to use ZeroGPU] from diffusers import StableDiffusionPipeline, EulerDiscreteScheduler, DPMSolverMultistepScheduler, UniPCMultistepScheduler, PNDMScheduler from ultralytics import YOLO from huggingface_hub import hf_hub_download import torch import os from PIL import Image, ImageFilter, ImageOps from huggingface_hub import login, hf_hub_download from gradio.themes import Default # For theming if "HF_TOKEN" in os.environ: login(os.environ["HF_TOKEN"]) else: raise ValueError("HF_TOKEN not found in environment variables. Please set it in Space settings.") device = "cuda" if torch.cuda.is_available() else "cpu" repo_id = "DreamingOracle/Quagmaform_alpha-1" filename = "DPS_Quagmaform_Alpha1.safetensors" model_path = hf_hub_download(repo_id=repo_id, filename=filename) if torch.cuda.is_available(): torch_dtype = torch.float16 else: torch_dtype = torch.float32 pipe = StableDiffusionPipeline.from_single_file(model_path, torch_dtype=torch_dtype) pipe = pipe.to(device) # Scheduler mapping (name to class) SCHEDULERS = { "PNDM": PNDMScheduler, "Euler": EulerDiscreteScheduler, "DPM++ 2M Karras": DPMSolverMultistepScheduler, "UniPC": UniPCMultistepScheduler, } # Download ADetailer model if not present adetailer_model_path = "face_yolov8n.pt" if not os.path.exists(adetailer_model_path): hf_hub_download(repo_id="Bingsu/adetailer", filename="face_yolov8n.pt", local_dir=".") adetailer_model = YOLO(adetailer_model_path) MAX_SEED = np.iinfo(np.int32).max MAX_IMAGE_SIZE = 1024 # Purple theme custom_theme = Default(primary_hue="purple") def infer( prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, scheduler_name="PNDM", save_format="png", progress=gr.Progress(track_tqdm=True),): if randomize_seed: seed = random.randint(0, MAX_SEED) generator = torch.Generator().manual_seed(seed) # Set scheduler dynamically scheduler_class = SCHEDULERS.get(scheduler_name, PNDMScheduler) pipe.scheduler = scheduler_class.from_config(pipe.scheduler.config) image = pipe( prompt=prompt, negative_prompt=negative_prompt, guidance_scale=guidance_scale, num_inference_steps=num_inference_steps, width=width, height=height, generator=generator, ).images[0] # --------------------------- # ADetailer post-processing for face enhancement (with padding + soft blend) # --------------------------- try: results = adetailer_model(image) if results and len(results) and getattr(results[0], "boxes", None): for box in results[0].boxes: x1, y1, x2, y2 = map(int, box.xyxy[0]) w = max(1, x2 - x1) h = max(1, y2 - y1) pad = int(max(10, 0.18 * max(w, h))) x1p = max(0, x1 - pad) y1p = max(0, y1 - pad) x2p = min(image.width, x2 + pad) y2p = min(image.height, y2 + pad) face = image.crop((x1p, y1p, x2p, y2p)) fw, fh = face.size fw8 = max(8, (fw // 8) * 8) fh8 = max(8, (fh // 8) * 8) if (fw8, fh8) != (fw, fh): face = face.resize((fw8, fh8), Image.LANCZOS) mask = Image.new("L", face.size, 255) blur_radius = max(4, int(min(face.size) / 10)) paste_mask = mask.filter(ImageFilter.GaussianBlur(radius=blur_radius)) inpaint_result = pipe( prompt=prompt + ", high detail face", image=face, mask_image=mask, strength=0.45, num_inference_steps=20, guidance_scale=7.5, generator=generator ).images[0] if paste_mask.mode != "L": paste_mask = paste_mask.convert("L") image.paste(inpaint_result, (x1p, y1p), paste_mask) except Exception as e: print("ADetailer post-process failed:", e) output_path = f"generated_image.{save_format}" image.save(output_path, format=save_format.upper()) return image, seed examples = [ "photorealistic portrait of a young woman, cinematic rim lighting, soft golden hour backlight, detailed skin pores, realistic eyelashes, 85mm lens, shallow depth of field, ultra-detailed, high dynamic range, film grain, detailed, 8k", "head helmet portrait of a futuristic armored soldier, worn brushed metal armor with neon blue accents, realistic cloth under-armor, weathering and scratches, volumetric rim light, cinematic pose, high detail, photoreal", "arctic mountain in snow, insulated modules, panorama view, blowing snow, cold blue light, realistic snow accumulation, high detail",] # Updated CSS 12826 css = """ #col-container { margin: 0 auto; max-width: 640px;} #community-row {justify-content: center; gap: 30px;} """ with gr.Blocks(css=css, theme=custom_theme) as demo: with gr.Column(elem_id="col-container"): gr.Markdown("# DPS-Quagmaform AI txt2img") 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", visible=True, ) 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=768, ) 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=10.0, step=0.1, value=5, ) num_inference_steps = gr.Slider( label="Number of inference steps", minimum=1, maximum=50, step=1, value=22, ) scheduler = gr.Dropdown( label="Sampler/Scheduler", choices=list(SCHEDULERS.keys()), value="PNDM", info="Change this setting for better quality in some situations" ) save_format = gr.Dropdown( choices=["png", "jpg"], value="png", label="Select Output Format" ) gr.Examples(examples=examples, inputs=[prompt]) # Community gr.Markdown("### Community") with gr.Row(elem_id="community-row"): gr.Button("Join Discord 💬", link="https://discord.gg/deepspace", variant="primary") gr.Button("Telegram En Español 📱", link="https://t.me/DeepSpaceHispano", variant="primary") gr.on( triggers=[run_button.click, prompt.submit], fn=infer, inputs=[ prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, scheduler, save_format, ], outputs=[result, seed], ) if __name__ == "__main__": demo.launch()