1st version

app.py

@@ -1,154 +1,181 @@
+import torch
+import spaces
 import gradio as gr
-import numpy as np
-import random
-
-# import spaces #[uncomment to use ZeroGPU]
 from diffusers import DiffusionPipeline
-import torch
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model_repo_id = "stabilityai/sdxl-turbo"  # Replace to the model you would like to use
+MAX_SEED = 2**32 - 1
 
-if torch.cuda.is_available():
-    torch_dtype = torch.float16
-else:
-    torch_dtype = torch.float32
+# 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,
+)
+pipe.to("cuda")
 
-pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
-pipe = pipe.to(device)
+'# ======== AoTI compilation + FA3 ======== (disabled on HF to avoid outdated AOTI/FA3 package errors)'
+# pipe.transformer.layers._repeated_blocks = ["ZImageTransformerBlock"]
+# spaces.aoti_blocks_load(pipe.transformer.layers, "zerogpu-aoti/Z-Image", variant="fa3")
 
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
+print("Pipeline loaded!")
 
-
-# @spaces.GPU #[uncomment to use ZeroGPU]
-def infer(
+@spaces.GPU
+def generate_image(
     prompt,
     negative_prompt,
-    seed,
-    randomize_seed,
-    width,
     height,
-    guidance_scale,
+    width,
     num_inference_steps,
+    guidance_scale,
+    seed,
+    randomize_seed,
     progress=gr.Progress(track_tqdm=True),
 ):
+    """Generate 4 images with seeds: seed, 2x, 3x, 4x (mod MAX_SEED)."""
     if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
+        seed = torch.randint(0, MAX_SEED, (1,)).item()
 
-    generator = torch.Generator().manual_seed(seed)
+    base_seed = int(seed) % MAX_SEED
+    if base_seed < 0:
+        base_seed += MAX_SEED
 
-    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]
+    seeds = [(base_seed * i) % MAX_SEED for i in range(1, 5)]
 
-    return image, seed
+    neg_prompt = None
+    if isinstance(negative_prompt, str) and negative_prompt.strip():
+        neg_prompt = negative_prompt
 
+    images = []
+    for s in seeds:
+        generator = torch.Generator("cuda").manual_seed(int(s))
+        image = pipe(
+            prompt=prompt,
+            negative_prompt=neg_prompt,
+            height=int(height),
+            width=int(width),
+            num_inference_steps=int(num_inference_steps),
+            guidance_scale=float(guidance_scale),  # 0.0 is recommended default for Turbo
+            generator=generator,
+        ).images[0]
+        images.append(image)
 
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
-]
+    return images, ", ".join(str(s) for s in seeds)
 
-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(" # Text-to-Image Gradio Template")
+# Example prompts
+examples = [
+    ["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"],
+]
 
-        with gr.Row():
-            prompt = gr.Text(
+# Build the Gradio interface
+with gr.Blocks(title="Z-Image-Turbo Demo") as demo:
+    gr.Markdown(
+        """
+        # 🎨 Z-Image-Turbo Demo
+        
+        Generate high-quality images using the [Tongyi-MAI/Z-Image-Turbo](https://huggingface.co/Tongyi-MAI/Z-Image-Turbo) model.
+        This turbo model generates images in just 8 inference steps!
+        """
+    )
+    
+    with gr.Row():
+        with gr.Column(scale=1):
+            prompt = gr.Textbox(
                 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=False,
+                placeholder="Enter your image description...",
+                lines=4,
             )
 
-            seed = gr.Slider(
-                label="Seed",
-                minimum=0,
-                maximum=MAX_SEED,
-                step=1,
-                value=0,
+            negative_prompt = gr.Textbox(
+                label="Negative Prompt",
+                placeholder="Things you don't want in the image...",
+                lines=3,
             )
 
-            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,  # Replace with defaults that work for your model
-                )
-
                 height = gr.Slider(
+                    minimum=512,
+                    maximum=2048,
+                    value=1024,
+                    step=64,
                     label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
                 )
-
-            with gr.Row():
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0,  # Replace with defaults that work for your model
+                width = gr.Slider(
+                    minimum=512,
+                    maximum=2048,
+                    value=1024,
+                    step=64,
+                    label="Width",
                 )
-
+            
+            with gr.Row():
                 num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
                     minimum=1,
-                    maximum=50,
+                    maximum=20,
+                    value=9,
                     step=1,
-                    value=2,  # Replace with defaults that work for your model
+                    label="Inference Steps",
+                    info="9 steps results in 8 DiT forwards",
+                )
+
+            guidance_scale = gr.Slider(
+                minimum=0.0,
+                maximum=7.0,
+                value=0.0,
+                step=0.1,
+                label="CFG Guidance Scale",
+                info="0 = no CFG (recommended for Turbo models)",
+            )
+            
+            with gr.Row():
+                seed = gr.Number(
+                    label="Seed",
+                    value=42,
+                    precision=0,
                 )
+                randomize_seed = gr.Checkbox(
+                    label="Randomize Seed",
+                    value=False,
+                )
+            
+            generate_btn = gr.Button("🚀 Generate", variant="primary", size="lg")
+        
+        with gr.Column(scale=1):
+            output_images = gr.Gallery(
+                label="Generated Images",
+                columns=2,
+                rows=2,
+                preview=True,
+            )
+            used_seeds = gr.Textbox(
+                label="Seeds Used (base, 2x, 3x, 4x)",
+                interactive=False,
+            )
+    
+    gr.Markdown("### 💡 Example Prompts")
+    gr.Examples(
+        examples=examples,
+        inputs=[prompt],
+        cache_examples=False,
+    )
 
-        gr.Examples(examples=examples, inputs=[prompt])
-    gr.on(
-        triggers=[run_button.click, prompt.submit],
-        fn=infer,
-        inputs=[
-            prompt,
-            negative_prompt,
-            seed,
-            randomize_seed,
-            width,
-            height,
-            guidance_scale,
-            num_inference_steps,
-        ],
-        outputs=[result, seed],
+    gr.Markdown("Demo by [mrfakename](https://x.com/realmrfakename). Model by Alibaba. The model is licensed under Apache 2.0, you can use generated images commercially! Thanks to [multimodalart](https://huggingface.co/multimodalart) for the FA3 + AoTI enhancements/speedups")
+    
+    # Connect the generate button
+    generate_btn.click(
+        fn=generate_image,
+        inputs=[prompt, negative_prompt, height, width, num_inference_steps, guidance_scale, seed, randomize_seed],
+        outputs=[output_images, used_seeds],
+    )
+    
+    # Also allow generating by pressing Enter in the prompt box
+    prompt.submit(
+        fn=generate_image,
+        inputs=[prompt, negative_prompt, height, width, num_inference_steps, guidance_scale, seed, randomize_seed],
+        outputs=[output_images, used_seeds],
     )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(mcp_server=True, show_error=True)

requirements.txt

@@ -1,6 +1,5 @@
-accelerate
-diffusers
-invisible_watermark
-torch
+gradio
+git+https://github.com/huggingface/diffusers
 transformers
-xformers
+kernels
+gradio[mcp]

zimage_client.py

@@ -0,0 +1,59 @@
+import os
+from datetime import datetime
+
+from gradio_client import Client
+
+
+def save_images(images, used_seeds, output_dir="outputs"):
+    os.makedirs(output_dir, exist_ok=True)
+    timestamp = datetime.now().strftime("%Y%m%d-%H%M%S")
+
+    for index, image in enumerate(images):
+        filename = f"{timestamp}_img{index + 1}.png"
+        path = os.path.join(output_dir, filename)
+
+        # `image` may be a PIL.Image or a path-like object depending on Gradio.
+        if hasattr(image, "save"):
+            image.save(path)
+        else:
+            from PIL import Image
+
+            Image.open(image).save(path)
+
+        print(f"Saved: {path}")
+
+    print(f"Seeds used: {used_seeds}")
+
+
+def main():
+    # Space that runs the Z-Image UI; change if needed.
+    space_id = "xiaopeng/Z_Image_Turbo_Free"
+
+    client = Client(space_id)
+
+    prompt = "A beautiful portrait photo of a woman, 4k, highly detailed"
+    negative_prompt = ""
+    height = 1024
+    width = 1024
+    num_inference_steps = 9
+    guidance_scale = 0.0  # 0 = no CFG, as recommended for Turbo
+    seed = 42
+    randomize_seed = False
+
+    images, used_seeds = client.predict(
+        prompt,
+        negative_prompt,
+        height,
+        width,
+        num_inference_steps,
+        guidance_scale,
+        seed,
+        randomize_seed,
+        api_name="/generate_image",
+    )
+
+    save_images(images, used_seeds)
+
+
+if __name__ == "__main__":
+    main()