Qwen Image 2.1 Create

README.md

@@ -0,0 +1,13 @@
+---
+title: Qwen-Image-2.1 Create
+emoji: 🎨
+colorFrom: blue
+colorTo: purple
+sdk: gradio
+sdk_version: 6.27.0
+python_version: '3.12'
+app_file: app.py
+pinned: false
+---
+
+Private staging for Qwen Image 2.1 launch. Flip to public at release.

app.py

@@ -0,0 +1,146 @@
+import json
+import os
+import random
+import secrets
+from datetime import datetime, timezone
+
+import gradio as gr
+import spaces
+import torch
+
+# Private staging: directedbykobyperez/Qwen-Image-2.1-Create
+# Bucket: directedbykobyperez/Qwen-Image-2.1-community
+COMMUNITY_BUCKET = "directedbykobyperez/Qwen-Image-2.1-community"
+COMMUNITY_URL = "https://huggingface.co/buckets/directedbykobyperez/Qwen-Image-2.1-community"
+
+MODEL_ID = os.environ.get("MODEL_ID", "Qwen/Qwen-Image-2.1")
+OUT = "/tmp/qwen21_out"
+os.makedirs(OUT, exist_ok=True)
+
+# Official 2K aspect presets from Qwen/Qwen-Image-2.1 model card.
+SIZE_PRESETS = {
+    "1:1 (2048x2048)": (2048, 2048),
+    "16:9 (2752x1536)": (2752, 1536),
+    "9:16 (1536x2752)": (1536, 2752),
+    "4:3 (2400x1792)": (2400, 1792),
+    "3:4 (1792x2400)": (1792, 2400),
+    "3:2 (2528x1696)": (2528, 1696),
+    "2:3 (1696x2528)": (1696, 2528),
+    "1:1 fast (1024x1024)": (1024, 1024),
+    "16:9 fast (1344x768)": (1344, 768),
+    "9:16 fast (768x1344)": (768, 1344),
+}
+
+pipe = None
+
+
+def get_pipeline():
+    global pipe
+    if pipe is None:
+        from diffusers import QwenImage21Pipeline
+        print(f"Loading {MODEL_ID} (first run downloads weights)...")
+        pipe = QwenImage21Pipeline.from_pretrained(
+            MODEL_ID, torch_dtype=torch.bfloat16,
+        ).to("cuda" if torch.cuda.is_available() else "cpu")
+        print("Pipeline loaded.")
+    return pipe
+
+
+def _upload_to_community(png_path, meta_dict, image_id):
+    """Silently share images/<id>/<id>.png + meta.json to the community bucket. Never raises."""
+    token = os.environ.get("HF_TOKEN") or os.environ.get("HUGGINGFACE_TOKEN")
+    if not token:
+        return
+    try:
+        from huggingface_hub import HfFileSystem
+        fs = HfFileSystem(token=token)
+        meta_local = os.path.join(OUT, f"{image_id}_meta.json")
+        with open(meta_local, "w", encoding="utf-8") as f:
+            json.dump(meta_dict, f, ensure_ascii=False, indent=2)
+        base = f"buckets/{COMMUNITY_BUCKET}/images/{image_id}"
+        fs.put_file(png_path, f"{base}/{image_id}.png")
+        fs.put_file(meta_local, f"{base}/meta.json")
+        print(f"[community] uploaded {image_id} to bucket")
+    except Exception as e:
+        print(f"[community] upload failed: {e}")
+
+
+@spaces.GPU(duration=180)
+def generate(prompt, negative_prompt, size_preset, steps, seed, edit_image):
+    pipeline = get_pipeline()
+    if seed is None or seed < 1:
+        try:
+            seed = int(seed) if seed else random.randint(1, 10**6)
+        except Exception:
+            seed = random.randint(1, 10**6)
+    seed = int(seed)
+    width, height = SIZE_PRESETS.get(size_preset, (1024, 1024))
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    generator = torch.Generator(device).manual_seed(seed)
+
+    kwargs = dict(
+        prompt=prompt,
+        width=width, height=height,
+        num_inference_steps=int(steps),
+        generator=generator,
+    )
+    if negative_prompt and negative_prompt.strip():
+        kwargs["negative_prompt"] = negative_prompt.strip()
+    mode = "t2i"
+    if edit_image is not None:
+        from PIL import Image as PILImage
+        kwargs["image"] = PILImage.open(edit_image).convert("RGB")
+        mode = "edit"
+
+    image = pipeline(**kwargs).images[0]
+
+    image_id = secrets.token_hex(6)
+    png_path = os.path.join(OUT, f"{image_id}.png")
+    image.save(png_path)
+
+    words = (prompt or "").strip().replace("\n", " ").split()[:8]
+    title = " ".join(words).title()[:80] if words else f"Qwen Image {image_id[:6]}"
+    _upload_to_community(png_path, {
+        "id": image_id,
+        "title": title,
+        "prompt": prompt or "",
+        "negative_prompt": negative_prompt or "",
+        "caption": (prompt or "").strip(),
+        "width": width,
+        "height": height,
+        "steps": int(steps),
+        "seed": seed,
+        "mode": mode,
+        "model": MODEL_ID,
+        "image_file": f"{image_id}.png",
+        "created_at": datetime.now(timezone.utc).isoformat(),
+    }, image_id)
+
+    return png_path, png_path, f"seed={seed} | {width}x{height} | {steps} steps | {mode}"
+
+
+with gr.Blocks(title="Qwen Image 2.1 Create") as demo:
+    gr.Markdown("# Qwen Image 2.1 Create\nText-to-image + image editing. Every generation auto-shares (png + meta.json) to the community bucket.")
+    with gr.Row():
+        with gr.Column():
+            prompt = gr.Textbox(label="Prompt", value='A neon shop sign that reads "QWEN IMAGE 2.1", rainy night, reflections on wet pavement', lines=4)
+            negative = gr.Textbox(label="Negative prompt (optional)", value="", lines=2)
+            edit_image = gr.Image(label="Edit image (optional — leave empty for text-to-image)", type="filepath")
+            with gr.Row():
+                size_preset = gr.Dropdown(label="Size", choices=list(SIZE_PRESETS.keys()), value="1:1 fast (1024x1024)")
+                steps = gr.Number(label="Steps (40 = official default)", value=30, precision=0)
+                seed = gr.Number(label="Seed (0 = random)", value=0, precision=0)
+            btn = gr.Button("Generate image", variant="primary")
+            gr.HTML("<div style='text-align:center;margin:-12px 0 -8px;font-size:0.9rem;'>Don't forget to<a style='margin-left:5px;padding:0;' href='https://huggingface.co/Qwen/Qwen-Image-2.1' target='_blank'>like the model ❤️</a></div>")
+        with gr.Column():
+            out_img = gr.Image(label="Result", type="filepath")
+            out_file = gr.File(label="Download PNG")
+            info = gr.Textbox(label="Info")
+    btn.click(fn=generate, inputs=[prompt, negative, size_preset, steps, seed, edit_image], outputs=[out_img, out_file, info], queue=True)
+    gr.Markdown(
+        "---\nPowered by [Qwen-Image-2.1](https://huggingface.co/Qwen/Qwen-Image-2.1) · "
+        "🎨 **Community images:** every generation is auto-shared (png + meta.json) to the "
+        f"[Qwen-Image-2.1-community bucket]({COMMUNITY_URL})"
+    )
+
+demo.queue().launch()

requirements.txt

@@ -0,0 +1,10 @@
+spaces
+torch
+git+https://github.com/huggingface/diffusers
+transformers>=5.17
+accelerate
+safetensors
+gradio
+huggingface-hub
+Pillow
+sentencepiece