"""Gradio entry point for local use and Hugging Face ZeroGPU.""" import faulthandler faulthandler.enable() # Include Python stacks if native startup code crashes. import spaces # Must initialize ZeroGPU before importing torch/diffusers. import base64 import html import inspect from io import BytesIO import os import secrets import time import gradio as gr import numpy as np from PIL import Image from generate import CHECKPOINTS, MODEL_ID, MODEL_SIZES, ROOT, generate_images, load_model ZERO_GPU = os.environ.get("SPACES_ZERO_GPU", "").lower() in {"1", "true", "yes"} def gpu_unavailable(error): """Recognize allocation/quota/CUDA failures without masking model errors.""" # Spaces puts the stable error category in gr.Error.title, while the # message varies by authentication/subscription tier and quota type. title = str(getattr(error, "title", "")).lower() message = str(getattr(error, "message", str(error))).lower() if title in {"zerogpu quota exceeded", "zerogpu pending credits exceeded"}: return True quota_failure = ( ("quota" in message and ("exceeded" in message or "exhausted" in message)) or "space app has reached its gpu limit" in message or "exceeded your zerogpu runs limit" in message or "too many zerogpu credits allocated" in message ) return quota_failure or any(text in message for text in ( "no gpu was available", "no gpu available", "cuda out of memory", "cuda is not available", "no nvidia driver", "not compiled with cuda", "cuda driver", "cuda-capable device", "gpu task aborted", )) # ZeroGPU requires CUDA model placement before its worker is allocated. GPU_READY = False MPS_READY = False if ZERO_GPU: try: for model_id in MODEL_SIZES: load_model(device="cuda", model_id=model_id) GPU_READY = True except Exception as error: if not gpu_unavailable(error): raise print(f"GPU unavailable at startup; using CPU: {error}") else: import torch GPU_READY = torch.cuda.is_available() MPS_READY = hasattr(torch.backends, "mps") and torch.backends.mps.is_available() def parse_settings(seed, checkpoint_text): try: seed = int(str(seed).strip()) checkpoints = sorted(set(int(part.strip()) for part in checkpoint_text.split(","))) except (ValueError, TypeError): raise gr.Error("Enter an integer seed and comma-separated integer checkpoints.") if not 0 <= seed <= 2147483647: raise gr.Error("Seed must be between 0 and 2147483647.") if not 2 <= len(checkpoints) <= 12 or checkpoints[0] != 0 or checkpoints[-1] != 1000: raise gr.Error("Use 2–12 checkpoints between 0 and 1000, including both endpoints.") return seed, checkpoints BLANK_IMAGE = Image.new("RGB", (256, 256), "#e7e5dc") def checkpoint_slots(checkpoints, frames=None): """Keep every timestep in its own fixed slot, including pending frames.""" frames = frames or {} return [(frames.get(step, BLANK_IMAGE), f"t = {step}") for step in checkpoints] def gaussian_noise(seed, model_id=MODEL_ID): noise = np.random.default_rng(int(seed)).standard_normal((MODEL_SIZES[model_id], MODEL_SIZES[model_id], 3)) return Image.fromarray(np.round((noise.clip(-1, 1) + 1) * 127.5).astype(np.uint8)) def progress_text(completed=0, started=None): elapsed = time.monotonic() - started if started is not None else 0 rate = completed / elapsed if elapsed > 0 and completed else 0 def duration(seconds): if seconds is None: return "--:--" seconds = int(seconds) hours, seconds = divmod(seconds, 3600) minutes, seconds = divmod(seconds, 60) return (f"{hours:02d}:" if hours else "") + f"{minutes:02d}:{seconds:02d}" remaining = 0 if completed == 1000 else (1000 - completed) / rate if rate else None return (f"{completed}/1000 [{duration(elapsed)}<{duration(remaining)}, " f"{f'{rate:.2f}' if rate else '?'}it/s]") def diffuse(image, seed, checkpoint_text, model_id=MODEL_ID): seed, checkpoints = parse_settings(seed, checkpoint_text) if image is None: raise gr.Error("Upload an image first.") pixels = np.asarray(image.convert("RGB").resize((MODEL_SIZES[model_id], MODEL_SIZES[model_id])), dtype=np.float32) / 127.5 - 1 noise = np.random.default_rng(seed).standard_normal(pixels.shape) alpha_bars = np.cumprod(1 - np.linspace(0.0001, 0.02, 1000)) started = time.monotonic() frames = {0: image.convert("RGB").resize((MODEL_SIZES[model_id], MODEL_SIZES[model_id]))} final_noise = None yield checkpoint_slots(checkpoints, frames), None, progress_html(0, started) for step in checkpoints: alpha = 1 if step == 0 else alpha_bars[step - 1] sample = np.sqrt(alpha) * pixels + np.sqrt(1 - alpha) * noise frame = Image.fromarray(np.round((sample.clip(-1, 1) + 1) * 127.5).astype(np.uint8)) frames[step] = frame final_noise = frame yield checkpoint_slots(checkpoints, frames), final_noise if step == 1000 else None, progress_html(step, started) def stream_denoising(seed, checkpoints, initial_image, device, model_id=MODEL_ID): started = time.monotonic() frames = {1000: initial_image if initial_image is not None else gaussian_noise(seed, model_id)} yield checkpoint_slots(checkpoints, frames), progress_html(0, started) print('using', device) for step, frame in generate_images(seed, checkpoints, initial_image, device=device, model_id=model_id): if frame is not None: frames[step] = frame yield (checkpoint_slots(checkpoints, frames) if frame is not None else gr.skip(), progress_html(1000 - step, started)) @spaces.GPU(duration=int(os.environ.get("GPU_DURATION_SECONDS", "120"))) def denoise_gpu(seed, checkpoints, initial_image, model_id=MODEL_ID): yield from stream_denoising(seed, checkpoints, initial_image, "cuda", model_id) def denoise(seed, checkpoint_text, input_mode, forward_noise, model_id=MODEL_ID): seed, checkpoints = parse_settings(seed, checkpoint_text) initial_image = forward_noise if input_mode == "Forward noise" else None if input_mode == "Forward noise" and initial_image is None: raise gr.Error("Run Diffuse first to create forward noise.") if GPU_READY: try: yield from denoise_gpu(seed, checkpoints, initial_image, model_id) return except Exception as error: if not gpu_unavailable(error): raise fallback = "MPS" if MPS_READY else "CPU" print(f"CUDA unavailable; restarting on {fallback}: {error}") gr.Info(f"CUDA unavailable or quota exhausted. Running on {fallback}.") if MPS_READY: try: yield from stream_denoising(seed, checkpoints, initial_image, "mps", model_id) return except (RuntimeError, NotImplementedError) as error: message = str(error).lower() if "mps" not in message and "metal" not in message: raise print(f"MPS unavailable; restarting on CPU: {error}") gr.Info("MPS unavailable. Running on CPU.") yield from stream_denoising(seed, checkpoints, initial_image, "cpu", model_id) # Reuse the original page's palette, typography, header, and panel styling. PAGE_CSS = (ROOT / "styles.css").read_text() + """ .gradio-container { max-width:1440px !important; padding:0 6% !important; background:var(--paper) !important; color:var(--ink) !important; background-image:radial-gradient(#59644d12 .7px,transparent .7px) !important; background-size:8px 8px !important; font-family:var(--sans) !important; } /* Keep the brand close to the top and Configure close to its divider. */ .gradio-container > .main { padding-top:0 !important; } #lab-header { padding:0 !important; margin:0 !important; } #lab-header .html-container { padding:0 !important; margin:0 !important; } #lab-header .topbar { height:38px; align-items:flex-start; padding-top:2px; } #lab-header .brand { font-size:24px; line-height:30px; } #lab-header .topbar-right { padding-top:12px; } #lab-main { gap:22px; padding:0; } .gradio-container > .main { padding-bottom:0 !important; flex-grow:0 !important; } .gradio-container, .gradio-container > .wrap { min-height:0 !important; } .gradio-container > .wrap { flex-grow:0 !important; } .gradio-container footer { margin-top:0 !important; padding:4px 0 !important; min-height:0 !important; } .gradio-container .gap:has(> #lab-header) { gap:0 !important; } .lab-panel { border-radius:0 !important; gap:12px !important; } .lab-panel h1,.lab-panel h2 { color:var(--ink); } .lab-panel.control-panel { padding-top:8px !important; } .control-panel > .html-container, .control-panel > .block { margin-top:0 !important; } .control-panel h1 { margin:0 !important; } .lab-panel.process-panel { padding:12px 14px !important; gap:8px !important; } .config-row { display:flex !important; gap:0 !important; width:100%; max-width:none; flex-wrap:nowrap !important; } .config-label { flex:0 0 110px !important; min-width:110px !important; height:40px !important; min-height:40px !important; padding:0 12px !important; display:flex; align-items:center; justify-content:flex-start !important; opacity:1 !important; background:#f0f2eb !important; border:1px solid #cfd3c9 !important; border-radius:3px 0 0 3px !important; color:var(--green) !important; font:10px var(--mono) !important; } .config-field { flex:1 1 0 !important; min-width:0 !important; border:0 !important; padding:0 !important; background:transparent !important; } /* Gradio nests the textbox inside a label; undo the original page's .config-row label styles so that wrapper cannot become a second box. */ .config-field label { display:block !important; width:100% !important; min-width:0 !important; border:0 !important; padding:0 !important; margin:0 !important; background:transparent !important; } .config-field .input-container { width:100% !important; max-width:none !important; } .config-field textarea,.config-field input { width:100% !important; max-width:none !important; box-sizing:border-box !important; height:40px !important; min-height:40px !important; border:1px solid #cfd3c9 !important; border-radius:0 3px 3px 0 !important; padding:10px 12px !important; font:11px var(--mono) !important; background:white !important; white-space:pre !important; overflow-x:auto !important; resize:none !important; } /* The dropdown wrapper adds spacing that textboxes do not have. */ .config-row:has(#model-field) { column-gap:0 !important; } .config-row:has(#model-field) > .form { margin:0 !important; padding:0 !important; gap:0 !important; } #model-field { margin:0 !important; padding:0 !important; height:40px !important; min-height:40px !important; max-height:40px !important; } #model-field .wrap, #model-field .input-container { margin:0 !important; } #model-field .container { padding:0 !important; margin:0 !important; } #model-field .wrap { box-sizing:border-box !important; height:40px !important; min-height:40px !important; max-height:40px !important; border:1px solid #cfd3c9 !important; border-radius:0 3px 3px 0 !important; box-shadow:none !important; } #model-field .wrap-inner { height:38px !important; min-height:0 !important; padding:0 12px !important; gap:0 !important; } /* The dropdown input sits inside its own bordered wrapper. */ #model-field input { border:0 !important; border-radius:0 !important; margin:0 !important; padding:0 !important; background:transparent !important; box-shadow:none !important; height:38px !important; min-height:38px !important; } .process-button { flex:0 0 auto !important; min-width:76px !important; height:28px !important; min-height:28px !important; padding:0 12px !important; background:var(--green) !important; color:white !important; border-radius:3px !important; font:10px var(--mono) !important; } #apply-timesteps { height:40px !important; border-radius:0 3px 3px 0 !important; } .process-heading { margin-bottom:0; } .process-toolbar { align-items:center !important; gap:12px !important; } /* Three columns keep progress centered independently of the controls. */ .process-toolbar { display:grid !important; grid-template-columns:minmax(0,1fr) max-content minmax(0,1fr); align-items:center !important; } .process-toolbar > .process-button { justify-self:start; } .process-toolbar > .process-button:last-child { justify-self:end; } .progress-counter { justify-self:center; min-width:0 !important; width:max-content !important; max-width:100%; border:0 !important; background:transparent !important; padding:0 !important; } .progress-counter .html-container { padding:0 !important; } .progress-counter pre { margin:0; padding:3px 5px; color:var(--coral); background:#f5f4ef; border:1px solid #d6dacf; border-radius:3px; font:11px var(--mono); text-align:center; white-space:pre; overflow-x:auto; } .checkpoint-track { background:transparent !important; border:0 !important; padding:0 !important; } /* Gradio's auto-height galleries still impose a 450px minimum. */ .checkpoint-track .grid-wrap { overflow-x:auto !important; overflow-y:hidden !important; height:auto !important; min-height:0 !important; max-height:none !important; padding:0 !important; } .checkpoint-track .grid-container { height:auto !important; min-height:0 !important; } .checkpoint-track .grid-container { gap:10px !important; grid-auto-rows:auto !important; } .checkpoint-track .thumbnail-item { min-width:0 !important; aspect-ratio:1 !important; height:auto !important; } .checkpoint-track .thumbnail-item { border-radius:0 !important; background:#e7e5dc !important; } .checkpoint-track .caption { font:11px var(--mono) !important; color:var(--green) !important; } .checkpoint-row { position:relative; gap:0 !important; overflow:visible !important; } .checkpoint-preview-host { position:relative; height:0 !important; min-height:0 !important; padding:0 !important; margin:0 !important; border:0 !important; overflow:visible !important; z-index:1000; } .checkpoint-preview-host .html-container { padding:0 !important; overflow:visible !important; } .checkpoint-preview { position:absolute; bottom:8px; width:max-content; max-width:none; z-index:1000; border:2px solid #fbfaf6; background:#fbfaf6; box-shadow:0 8px 24px #25312852; } .checkpoint-preview summary { cursor:pointer; padding:3px 6px; font:10px var(--mono); } .checkpoint-preview img { width:auto; height:auto; max-width:none; display:block; } .checkpoint-preview:not([open]) { display:none; } #reverse-input { justify-self:end; width:max-content !important; min-width:0 !important; border:0 !important; background:transparent !important; padding:0 !important; } #reverse-input .wrap, #reverse-input .form, #reverse-input .container { width:max-content !important; min-width:0 !important; padding:0 !important; gap:6px !important; justify-content:flex-end !important; } #reverse-input label { flex:0 0 auto !important; width:max-content !important; margin:0 !important; padding:4px 6px !important; } @media(max-width:620px) { .gradio-container { padding:0 4% !important; } .config-label { flex-basis:82px !important; min-width:82px !important; } .process-heading { align-items:flex-start; flex-direction:column; } .process-toolbar { grid-template-columns:1fr !important; } .process-toolbar > :last-child { justify-self:end; } } """ PAGE_HEAD = """""" # Gradio 6 moved CSS/head from Blocks to launch; support both runtime versions. PAGE_OPTIONS = {"css": PAGE_CSS, "head": PAGE_HEAD} BLOCK_OPTIONS = PAGE_OPTIONS if "css" not in inspect.signature(gr.Blocks.launch).parameters else {} LAUNCH_OPTIONS = PAGE_OPTIONS if not BLOCK_OPTIONS else {} # Install after a native selection event so this also works when Spaces mounts # Blocks without using the local launch options. Traverse open shadow roots. INSTALL_PREVIEW_DISMISS = """() => { if (window.diffusionDismissInstalled) return; window.diffusionDismissInstalled = true; const closeOutside = (root, path) => { for (const preview of root.querySelectorAll(".checkpoint-preview[open]")) { if (!path.includes(preview)) preview.open = false; } for (const element of root.querySelectorAll("*")) { if (element.shadowRoot) closeOutside(element.shadowRoot, path); } }; document.addEventListener("pointerdown", event => { closeOutside(document, event.composedPath()); }, true); document.addEventListener("keydown", event => { if (event.key === "Escape") closeOutside(document, []); }); }""" def progress_html(completed=0, started=None): return "
" + progress_text(completed, started) + "
" def reset_tracks(seed, checkpoint_text, image, model_id=MODEL_ID): seed, steps = parse_settings(seed, checkpoint_text) initial = {0: image.convert("RGB").resize((MODEL_SIZES[model_id], MODEL_SIZES[model_id]))} if image is not None else {} return (", ".join(map(str, steps)), gr.Gallery(value=checkpoint_slots(steps, initial), columns=len(steps), allow_preview=False, interactive=False), gr.Gallery(value=checkpoint_slots(steps, {1000: gaussian_noise(seed, model_id)}), columns=len(steps), allow_preview=False, interactive=False), None, progress_html(), progress_html()) def change_source(filepath, seed, checkpoint_text, model_id=MODEL_ID): with Image.open(filepath) as uploaded: image = uploaded.convert("RGB") return (image, *reset_tracks(seed, checkpoint_text, image, model_id)) def change_model(seed, checkpoint_text, image, model_id): source_paths = { MODEL_ID: ROOT / "assets/celeba_hq_validation_91_256.png", "google/ddpm-cifar10-32": ROOT / "assets/cifar10_train_23_32.png", } return change_source(source_paths[model_id], seed, checkpoint_text, model_id) def show_checkpoint_preview(gallery, event: gr.SelectData): if not gallery: return "" index = int(event.index) image, caption = gallery[index] if isinstance(image, dict): image = image["path"] if not isinstance(image, Image.Image): with Image.open(image) as loaded: image = loaded.convert("RGB") buffer = BytesIO() image.save(buffer, format="PNG") encoded = base64.b64encode(buffer.getvalue()).decode("ascii") width, height = image.size preview_width = width + 4 center = 100 * (index + 0.5) / len(gallery) # Force fresh markup even when the same thumbnail is clicked again, so # a details element closed in the browser is recreated in its open state. preview_id = secrets.token_hex(8) return (f'
' 'Close preview' f'{html.escape(caption or ' '
') with gr.Blocks(title="Diffusion — A DDPM in motion", **BLOCK_OPTIONS) as demo: gr.HTML('
' '' 'diffusion model
' '
', elem_id="lab-header", container=False) with gr.Column(elem_id="lab-main"): with gr.Column(elem_classes=["lab-panel", "control-panel"]): gr.HTML("

Configure

") with gr.Row(elem_classes="config-row"): gr.Button("Model", interactive=False, scale=0, min_width=110, elem_classes="config-label") model = gr.Dropdown(choices=list(MODEL_SIZES), value=MODEL_ID, scale=10, min_width=0, show_label=False, container=False, elem_id="model-field", elem_classes="config-field") with gr.Row(elem_classes="config-row"): gr.Button("Timesteps", interactive=False, scale=0, min_width=110, elem_classes="config-label") checkpoints = gr.Textbox(value=", ".join(map(str, CHECKPOINTS)), show_label=False, scale=10, min_width=0, lines=1, max_lines=1, container=False, elem_classes="config-field") apply_button = gr.Button("Apply", scale=0, min_width=76, elem_id="apply-timesteps", elem_classes="process-button") with gr.Row(elem_classes="config-row"): randomize = gr.Button("Seed", scale=0, min_width=110, elem_classes="config-label") seed = gr.Textbox(value="42", show_label=False, scale=10, min_width=0, lines=1, max_lines=1, container=False, elem_classes="config-field") randomize.click(lambda: str(secrets.randbelow(2**31)), outputs=seed, queue=False) default_image = ROOT / "assets/celeba_hq_validation_91_256.png" initial_image = Image.open(default_image).convert("RGB") if default_image.exists() else None initial_frames = {0: initial_image.resize((256, 256))} if initial_image is not None else {} source = gr.State(initial_image) forward_noise = gr.State(None) with gr.Column(elem_classes=["lab-panel", "process-panel"]): gr.HTML('

Forward process: image to noise

' 'xt = √α̅t xt−1' ' + √(1 − α̅t) εt
') with gr.Row(elem_classes="process-toolbar"): diffuse_button = gr.Button("Diffuse", elem_classes="process-button") forward_status = gr.HTML(value=progress_html(), container=False, elem_classes="progress-counter") upload_source = gr.UploadButton("Change image", file_types=["image"], type="filepath", elem_classes="process-button") with gr.Column(elem_classes="checkpoint-row"): forward_preview = gr.HTML("", container=False, elem_classes="checkpoint-preview-host") forward_gallery = gr.Gallery(value=checkpoint_slots(CHECKPOINTS, initial_frames), show_label=False, columns=len(CHECKPOINTS), rows=1, height="auto", object_fit="cover", allow_preview=False, interactive=False, elem_id="forward-images", elem_classes="checkpoint-track") forward_gallery.select(show_checkpoint_preview, [forward_gallery], [forward_preview], queue=False, show_progress="hidden").then( fn=None, js=INSTALL_PREVIEW_DISMISS, queue=False) diffuse_button.click(diffuse, [source, seed, checkpoints, model], [forward_gallery, forward_noise, forward_status]) with gr.Column(elem_classes=["lab-panel", "process-panel"]): gr.HTML('

Reverse process: noise to image

' '

xt−1 = 1/√αt' ' (xt − βt/√(1−α̅t)' ' εθ(xt,t)) + σtz

') with gr.Row(elem_classes="process-toolbar"): denoise_button = gr.Button("Denoise", elem_classes="process-button") status = gr.HTML(value=progress_html(), container=False, elem_classes="progress-counter") input_mode = gr.Radio(["Gaussian noise", "Forward noise"], value="Gaussian noise", show_label=False, container=False, scale=0, min_width=0, elem_id="reverse-input") with gr.Column(elem_classes="checkpoint-row"): reverse_preview = gr.HTML("", container=False, elem_classes="checkpoint-preview-host") reverse_gallery = gr.Gallery(value=checkpoint_slots(CHECKPOINTS, {1000: gaussian_noise(42)}), show_label=False, columns=len(CHECKPOINTS), rows=1, height="auto", object_fit="cover", allow_preview=False, interactive=False, elem_classes="checkpoint-track") reverse_gallery.select(show_checkpoint_preview, [reverse_gallery], [reverse_preview], queue=False, show_progress="hidden").then( fn=None, js=INSTALL_PREVIEW_DISMISS, queue=False) denoise_button.click(denoise, [seed, checkpoints, input_mode, forward_noise, model], [reverse_gallery, status], concurrency_limit=1, **({"stream_every": 0} if "stream_every" in inspect.signature(denoise_button.click).parameters else {})) def clear_previews(): return "", "" for control, event_name in [(apply_button, "click"), (upload_source, "upload"), (diffuse_button, "click"), (denoise_button, "click")]: getattr(control, event_name)(clear_previews, outputs=[forward_preview, reverse_preview], queue=False, show_progress="hidden") reset_outputs = [checkpoints, forward_gallery, reverse_gallery, forward_noise, forward_status, status] apply_button.click(reset_tracks, [seed, checkpoints, source, model], reset_outputs, queue=False) model.change(change_model, [seed, checkpoints, source, model], [source, *reset_outputs]) model.change(clear_previews, outputs=[forward_preview, reverse_preview], queue=False) upload_source.upload(change_source, [upload_source, seed, checkpoints, model], [source, *reset_outputs], queue=False) if __name__ == "__main__": demo.queue().launch(**LAUNCH_OPTIONS)