Convert app to gr.Workflow calling Z-Image-Turbo via HF Inference API

Replace the local-diffusers/GPU Blocks app with a gr.Workflow pipeline:
Prompt (reference) -> Z-Image-Turbo (model operator, fal-ai provider)
-> Output Image (subject). Calls InferenceClient.text_to_image on
Tongyi-MAI/Z-Image-Turbo, so no GPU or local weights are needed.

- app.py: gr.Workflow(graph=workflow.json, bind={"text_to_image": ...})
- workflow.json: pre-wired schema-v2 canvas (model operator + edges)
- requirements.txt: drop diffusers/torch/kernels, add huggingface_hub
- README.md: pin gradio 6.1.0 (ships gr.Workflow), set hf_oauth: true

README.md

@@ -4,9 +4,64 @@ emoji: 🖼️
 colorFrom: yellow
 colorTo: yellow
 sdk: gradio
-sdk_version: 6.0.1
+sdk_version: 6.1.0
 app_file: app.py
 pinned: true
+hf_oauth: true
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Z-Image-Turbo (Gradio Workflow)
+
+A visual, node-based image-generation app built with `gr.Workflow`. It calls
+the [Tongyi-MAI/Z-Image-Turbo](https://huggingface.co/Tongyi-MAI/Z-Image-Turbo)
+model through the Hugging Face Inference API (served by the `fal-ai` provider)
+— no GPU or local weights required.
+
+## How it works
+
+The workflow (defined in [`workflow.json`](./workflow.json)) is three nodes:
+
+1. **Prompt** (reference) — the text prompt you want to render.
+2. **Z-Image-Turbo** (operator, `kind: "model"`) — calls the model via
+   `InferenceClient.text_to_image`.
+3. **Output Image** (subject) — the generated image, also exposed as the
+   `/output_image` API endpoint.
+
+Edit the topology on the canvas (drag nodes, change the prompt, rewire) and
+hit **Run**. Changes are saved back to `workflow.json`.
+
+## Running locally
+
+```bash
+pip install -r requirements.txt
+huggingface_hub login   # provides the HF token used by the InferenceClient
+python app.py
+```
+
+Open the **write-access link** printed at launch to edit the workflow; plain
+local/share URLs open it read-only.
+
+## Deploying
+
+This is a standard Gradio app — deploy it with:
+
+```bash
+gradio deploy
+```
+
+`hf_oauth: true` is set so that, on a Space, each visitor signs in with their
+own HF account and the model call runs under their own token/quota. The Space
+owner can edit and save the workflow; visitors get a read-only view and can
+run the pipeline.
+
+## API access
+
+Every Workflow app is a Gradio app, so it exposes a REST endpoint per output
+(subject) node — e.g. `/output_image`:
+
+```python
+from gradio_client import Client
+
+client = Client("your-username/your-space")
+client.view_api()  # list endpoints and parameters
+```

app.py

@@ -1,260 +1,38 @@
-import torch
-import spaces
-import gradio as gr
-from diffusers import DiffusionPipeline
+import os
 
-# Load the pipeline once at startup
-print("Loading Z-Image-Turbo pipeline...")
-pipe = DiffusionPipeline.from_pretrained(
-    "Tongyi-MAI/Z-Image-Turbo",
-    torch_dtype=torch.bfloat16,
-    low_cpu_mem_usage=False,
+import gradio as gr
+from huggingface_hub import InferenceClient
+
+# Z-Image-Turbo is served through the Hugging Face Inference API via the
+# fal-ai provider. The token is provided by the Space (HF_TOKEN secret) or,
+# locally, by `huggingface_hub login`. On a Space with `hf_oauth: true`,
+# the workflow canvas forwards each visitor's own OAuth token to the model
+# call, so they run inference under their own account/quota.
+_HF_TOKEN = os.environ.get("HF_TOKEN")
+_client = InferenceClient(
+    provider="fal-ai",
+    api_key=_HF_TOKEN,
+    model="Tongyi-MAI/Z-Image-Turbo",
 )
-pipe.to("cuda")
 
-# ======== AoTI compilation + FA3 ========
-# pipe.transformer.layers._repeated_blocks = ["ZImageTransformerBlock"]
-# spaces.aoti_blocks_load(pipe.transformer.layers, "zerogpu-aoti/Z-Image", variant="fa3")
 
-print("Pipeline loaded!")
+def text_to_image(prompt: str):
+    """Generate an image from a text prompt using Z-Image-Turbo.
 
-@spaces.GPU
-def generate_image(prompt, height, width, num_inference_steps, seed, randomize_seed, progress=gr.Progress(track_tqdm=True)):
-    """Generate an image from the given prompt."""
-    if randomize_seed:
-        seed = torch.randint(0, 2**32 - 1, (1,)).item()
-    
-    generator = torch.Generator("cuda").manual_seed(int(seed))
-    image = pipe(
-        prompt=prompt,
-        height=int(height),
-        width=int(width),
-        num_inference_steps=int(num_inference_steps),
-        guidance_scale=0.0,
-        generator=generator,
-    ).images[0]
-    
-    return image, seed
+    Exposed as a bound node on the workflow canvas and as the workflow's
+    `/text_to_image` API endpoint (the subject node name derives from this).
+    Guidance is fixed at 0 (the Turbo models use no CFG) and the model
+    defaults to ~8 effective denoising steps.
+    """
+    if not prompt or not prompt.strip():
+        raise gr.Error("Please enter a prompt.")
+    return _client.text_to_image(prompt, model="Tongyi-MAI/Z-Image-Turbo")
 
-# Example prompts
-examples = [
-    ["Young Chinese woman in red Hanfu, intricate embroidery. Impeccable makeup, red floral forehead pattern. Elaborate high bun, golden phoenix headdress, red flowers, beads. Holds round folding fan with lady, trees, bird. Neon lightning-bolt lamp, bright yellow glow, above extended left palm. Soft-lit outdoor night background, silhouetted tiered pagoda, blurred colorful distant lights."],
-    ["A majestic dragon soaring through clouds at sunset, scales shimmering with iridescent colors, detailed fantasy art style"],
-    ["Cozy coffee shop interior, warm lighting, rain on windows, plants on shelves, vintage aesthetic, photorealistic"],
-    ["Astronaut riding a horse on Mars, cinematic lighting, sci-fi concept art, highly detailed"],
-    ["Portrait of a wise old wizard with a long white beard, holding a glowing crystal staff, magical forest background"],
-]
 
-# Custom theme with modern aesthetics (Gradio 6)
-custom_theme = gr.themes.Soft(
-    primary_hue="yellow",
-    secondary_hue="amber",
-    neutral_hue="slate",
-    font=gr.themes.GoogleFont("Inter"),
-    text_size="lg",
-    spacing_size="md",
-    radius_size="lg"
-).set(
-    button_primary_background_fill="*primary_500",
-    button_primary_background_fill_hover="*primary_600",
-    block_title_text_weight="600",
+demo = gr.Workflow(
+    graph="workflow.json",
+    bind={"text_to_image": text_to_image},
 )
 
-# Build the Gradio interface
-with gr.Blocks(fill_height=True) as demo:
-    # Header
-    gr.Markdown(
-        """
-        # 🎨 Z-Image-Turbo
-        **Ultra-fast AI image generation** • Generate stunning images in just 8 steps
-        """,
-        elem_classes="header-text"
-    )
-    
-    with gr.Row(equal_height=False):
-        # Left column - Input controls
-        with gr.Column(scale=1, min_width=320):
-            prompt = gr.Textbox(
-                label="✨ Your Prompt",
-                placeholder="Describe the image you want to create...",
-                lines=5,
-                max_lines=10,
-                autofocus=True,
-            )
-            
-            with gr.Accordion("⚙️ Advanced Settings", open=False):
-                with gr.Row():
-                    height = gr.Slider(
-                        minimum=512,
-                        maximum=2048,
-                        value=1024,
-                        step=64,
-                        label="Height",
-                        info="Image height in pixels"
-                    )
-                    width = gr.Slider(
-                        minimum=512,
-                        maximum=2048,
-                        value=1024,
-                        step=64,
-                        label="Width",
-                        info="Image width in pixels"
-                    )
-                
-                num_inference_steps = gr.Slider(
-                    minimum=1,
-                    maximum=20,
-                    value=9,
-                    step=1,
-                    label="Inference Steps",
-                    info="9 steps = 8 DiT forwards (recommended)"
-                )
-                
-                with gr.Row():
-                    randomize_seed = gr.Checkbox(
-                        label="🎲 Random Seed",
-                        value=True,
-                    )
-                    seed = gr.Number(
-                        label="Seed",
-                        value=42,
-                        precision=0,
-                        visible=False,
-                    )
-                
-                def toggle_seed(randomize):
-                    return gr.Number(visible=not randomize)
-                
-                randomize_seed.change(
-                    toggle_seed,
-                    inputs=[randomize_seed],
-                    outputs=[seed]
-                )
-            
-            generate_btn = gr.Button(
-                "🚀 Generate Image",
-                variant="primary",
-                size="lg",
-                scale=1
-            )
-            
-            # Example prompts
-            gr.Examples(
-                examples=examples,
-                inputs=[prompt],
-                label="💡 Try these prompts",
-                examples_per_page=5,
-            )
-        
-        # Right column - Output
-        with gr.Column(scale=1, min_width=320):
-            output_image = gr.Image(
-                label="Generated Image",
-                type="pil",
-                format="png",
-                show_label=False,
-                height=600,
-                buttons=["download", "share"],
-            )
-            
-            used_seed = gr.Number(
-                label="🎲 Seed Used",
-                interactive=False,
-                container=True,
-            )
-    
-    # Footer credits
-    gr.Markdown(
-        """
-        ---
-        <div style="text-align: center; opacity: 0.7; font-size: 0.9em; margin-top: 1rem;">
-        <strong>Model:</strong> <a href="https://huggingface.co/Tongyi-MAI/Z-Image-Turbo" target="_blank">Tongyi-MAI/Z-Image-Turbo</a> (Apache 2.0 License) • 
-        <strong>Demo by:</strong> <a href="https://x.com/realmrfakename" target="_blank">@mrfakename</a> • 
-        <strong>Redesign by:</strong> AnyCoder • 
-        <strong>Optimizations:</strong> <a href="https://huggingface.co/multimodalart" target="_blank">@multimodalart</a> (FA3 + AoTI)
-        </div>
-        """,
-        elem_classes="footer-text"
-    )
-    
-    # Connect the generate button
-    generate_btn.click(
-        fn=generate_image,
-        inputs=[prompt, height, width, num_inference_steps, seed, randomize_seed],
-        outputs=[output_image, used_seed],
-    )
-    
-    # Also allow generating by pressing Enter in the prompt box
-    prompt.submit(
-        fn=generate_image,
-        inputs=[prompt, height, width, num_inference_steps, seed, randomize_seed],
-        outputs=[output_image, used_seed],
-    )
-
 if __name__ == "__main__":
-    demo.launch(
-        theme=custom_theme,
-        css="""
-        .header-text h1 {
-            font-size: 2.5rem !important;
-            font-weight: 700 !important;
-            margin-bottom: 0.5rem !important;
-            background: linear-gradient(135deg, #fbbf24 0%, #f59e0b 100%);
-            -webkit-background-clip: text;
-            -webkit-text-fill-color: transparent;
-            background-clip: text;
-        }
-        
-        .header-text p {
-            font-size: 1.1rem !important;
-            color: #64748b !important;
-            margin-top: 0 !important;
-        }
-        
-        .footer-text {
-            padding: 1rem 0;
-        }
-        
-        .footer-text a {
-            color: #f59e0b !important;
-            text-decoration: none !important;
-            font-weight: 500;
-        }
-        
-        .footer-text a:hover {
-            text-decoration: underline !important;
-        }
-        
-        /* Mobile optimizations */
-        @media (max-width: 768px) {
-            .header-text h1 {
-                font-size: 1.8rem !important;
-            }
-            
-            .header-text p {
-                font-size: 1rem !important;
-            }
-        }
-        
-        /* Smooth transitions */
-        button, .gr-button {
-            transition: all 0.2s ease !important;
-        }
-        
-        button:hover, .gr-button:hover {
-            transform: translateY(-1px);
-            box-shadow: 0 4px 12px rgba(0, 0, 0, 0.15) !important;
-        }
-        
-        /* Better spacing */
-        .gradio-container {
-            max-width: 1400px !important;
-            margin: 0 auto !important;
-        }
-        """,
-        footer_links=[
-            "api",
-            "gradio"
-        ],
-        mcp_server=True
-    )
+    demo.launch()

requirements.txt

@@ -1,5 +1,3 @@
-gradio
-git+https://github.com/huggingface/diffusers
-transformers
-kernels
-gradio[mcp]
+gradio>=6.1
+huggingface_hub>=0.36.0
+gradio[mcp]

workflow.json

@@ -0,0 +1,88 @@
+{
+  "schema_version": "2",
+  "name": "Z-Image-Turbo",
+  "description": "Ultra-fast AI image generation with Z-Image-Turbo via the Hugging Face Inference API (fal-ai provider).",
+  "runtime": { "default": "client" },
+  "view": { "default": "canvas" },
+  "references": [
+    {
+      "id": "ref_prompt",
+      "label": "Prompt",
+      "role": "reference",
+      "asset_type": "text",
+      "inputs": [
+        { "id": "in", "label": "Prompt", "type": "text" }
+      ],
+      "outputs": [
+        { "id": "out", "label": "Prompt", "type": "text" }
+      ],
+      "x": 80,
+      "y": 160,
+      "width": 240,
+      "height": 120,
+      "data": {
+        "out": "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
+      }
+    }
+  ],
+  "operators": [
+    {
+      "id": "op_zimage",
+      "label": "Z-Image-Turbo",
+      "role": "operator",
+      "kind": "model",
+      "source": "hf://Tongyi-MAI/Z-Image-Turbo",
+      "model_id": "Tongyi-MAI/Z-Image-Turbo",
+      "pipeline_tag": "text-to-image",
+      "provider": "fal-ai",
+      "inputs": [
+        { "id": "in_0", "label": "Prompt", "type": "text", "required": true }
+      ],
+      "outputs": [
+        { "id": "out_0", "label": "Image", "type": "image", "output_index": 0 }
+      ],
+      "x": 440,
+      "y": 150,
+      "width": 260,
+      "height": 130,
+      "data": {}
+    }
+  ],
+  "subjects": [
+    {
+      "id": "sub_image",
+      "label": "Output Image",
+      "role": "subject",
+      "asset_type": "image",
+      "inputs": [
+        { "id": "in", "label": "Image", "type": "image" }
+      ],
+      "outputs": [
+        { "id": "out", "label": "Image", "type": "image" }
+      ],
+      "x": 800,
+      "y": 160,
+      "width": 240,
+      "height": 120,
+      "data": {}
+    }
+  ],
+  "edges": [
+    {
+      "id": "e_prompt_to_model",
+      "from_node_id": "ref_prompt",
+      "from_port_id": "out",
+      "to_node_id": "op_zimage",
+      "to_port_id": "in_0",
+      "type": "text"
+    },
+    {
+      "id": "e_model_to_output",
+      "from_node_id": "op_zimage",
+      "from_port_id": "out_0",
+      "to_node_id": "sub_image",
+      "to_port_id": "in",
+      "type": "image"
+    }
+  ]
+}