Create app.py

app.py

@@ -0,0 +1,263 @@
+import os
+import gradio as gr
+import numpy as np
+import random
+import spaces
+import torch
+from diffusers.pipelines.glm_image import GlmImagePipeline
+from PIL import Image
+
+dtype = torch.bfloat16
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+MAX_SEED = np.iinfo(np.int32).max
+MAX_IMAGE_SIZE = 2048
+
+# Load model
+pipe = GlmImagePipeline.from_pretrained(
+    "zai-org/GLM-Image",
+    torch_dtype=torch.bfloat16,
+    device_map="cuda"
+)
+
+
+def get_duration(prompt, input_images, width, height, num_inference_steps, guidance_scale, seed, progress=gr.Progress(track_tqdm=True)):
+    """Calculate GPU duration based on parameters"""
+    num_images = 0 if input_images is None else len(input_images)
+    step_duration = 1.2 + 0.5 * num_images
+    return max(90, num_inference_steps * step_duration + 15)
+
+
+@spaces.GPU(duration=get_duration)
+def generate_image(prompt, input_images, width, height, num_inference_steps, guidance_scale, seed, progress=gr.Progress(track_tqdm=True)):
+    """Generate image using GLM-Image pipeline"""
+    generator = torch.Generator(device="cuda").manual_seed(seed)
+    
+    # Prepare image list for image-to-image mode
+    image_list = None
+    if input_images is not None and len(input_images) > 0:
+        image_list = []
+        for item in input_images:
+            img = item[0] if isinstance(item, tuple) else item
+            if isinstance(img, str):
+                img = Image.open(img).convert("RGB")
+            elif isinstance(img, Image.Image):
+                img = img.convert("RGB")
+            image_list.append(img)
+    
+    pipe_kwargs = {
+        "prompt": prompt,
+        "height": height,
+        "width": width,
+        "num_inference_steps": num_inference_steps,
+        "guidance_scale": guidance_scale,
+        "generator": generator,
+    }
+    
+    # Add images for image-to-image mode
+    if image_list is not None:
+        pipe_kwargs["image"] = image_list
+    
+    if progress:
+        progress(0, desc="Starting generation...")
+    
+    image = pipe(**pipe_kwargs).images[0]
+    return image
+
+
+def infer(prompt, input_images=None, seed=42, randomize_seed=False, width=1024, height=1024, 
+          num_inference_steps=50, guidance_scale=1.5, progress=gr.Progress(track_tqdm=True)):
+    """Main inference function"""
+    
+    if randomize_seed:
+        seed = random.randint(0, MAX_SEED)
+    
+    # Ensure dimensions are multiples of 32
+    width = (width // 32) * 32
+    height = (height // 32) * 32
+    
+    progress(0.1, desc="Preparing generation...")
+    
+    image = generate_image(
+        prompt,
+        input_images,
+        width,
+        height,
+        num_inference_steps,
+        guidance_scale,
+        seed,
+        progress
+    )
+    
+    return image, seed
+
+
+def update_dimensions_from_image(image_list):
+    """Update width/height sliders based on uploaded image aspect ratio.
+    Keeps dimensions proportional with both sides as multiples of 32."""
+    if image_list is None or len(image_list) == 0:
+        return 1024, 1024  # Default dimensions
+    
+    # Get the first image to determine dimensions
+    item = image_list[0]
+    img = item[0] if isinstance(item, tuple) else item
+    
+    if isinstance(img, str):
+        img = Image.open(img)
+    
+    img_width, img_height = img.size
+    aspect_ratio = img_width / img_height
+    
+    if aspect_ratio >= 1:  # Landscape or square
+        new_width = 1024
+        new_height = int(1024 / aspect_ratio)
+    else:  # Portrait
+        new_height = 1024
+        new_width = int(1024 * aspect_ratio)
+    
+    # Round to nearest multiple of 32 (GLM-Image requirement)
+    new_width = round(new_width / 32) * 32
+    new_height = round(new_height / 32) * 32
+    
+    # Ensure within valid range
+    new_width = max(256, min(MAX_IMAGE_SIZE, new_width))
+    new_height = max(256, min(MAX_IMAGE_SIZE, new_height))
+    
+    return new_width, new_height
+
+
+# Text-to-Image examples
+examples_t2i = [
+    ["A beautifully designed modern food magazine style dessert recipe illustration, themed around a raspberry mousse cake. Clean and bright layout with a bold black title 'Raspberry Mousse Cake Recipe Guide', soft-lit close-up of the finished cake showing light pink layers with fresh raspberries and mint leaves."],
+    ["A kawaii die-cut sticker of a chubby orange cat, featuring big sparkly eyes and a happy smile with paws raised in greeting and a heart-shaped pink nose. Smooth rounded lines with black outlines and soft gradient shading with pink cheeks."],
+    ["Soaking wet capybara taking shelter under a banana leaf in the rainy jungle, close up photo, water droplets visible on fur, lush green background"],
+    ["A photorealistic portrait of an astronaut in a detailed spacesuit, floating in space with Earth visible in the background, dramatic lighting from the sun"],
+    ["An elegant art deco poster design for a jazz concert, featuring geometric patterns in gold and deep blue, stylized saxophone silhouette, vintage typography"],
+]
+
+# Image-to-Image examples
+examples_i2i = [
+    ["Transform this scene into a cyberpunk cityscape at night with neon lights and flying cars"],
+    ["Replace the background with an underwater coral reef scene with tropical fish"],
+    ["Convert this to a watercolor painting style with soft pastel colors"],
+]
+
+css = """
+#col-container {
+    margin: 0 auto;
+    max-width: 1200px;
+}
+.gallery-container img {
+    object-fit: contain;
+}
+"""
+
+with gr.Blocks() as demo:
+    
+    with gr.Column(elem_id="col-container"):
+        gr.Markdown("""# GLM-Image
+[[Model](https://huggingface.co/zai-org/GLM-Image)]
+        """)
+        
+        with gr.Row():
+            with gr.Column():
+                prompt = gr.Text(
+                    label="Prompt",
+                    show_label=False,
+                    max_lines=4,
+                    placeholder="Enter your prompt (for text-to-image) or editing instructions (for image-to-image)",
+                    container=False,
+                    scale=3
+                )
+                
+                run_button = gr.Button("🎨 Generate", variant="primary", scale=1)
+                
+                with gr.Accordion("📷 Input Image(s) (optional - for image-to-image mode)", open=True):
+                    input_images = gr.Gallery(
+                        label="Input Image(s)",
+                        type="pil",
+                        columns=3,
+                        rows=1,
+                        elem_classes="gallery-container"
+                    )
+                    gr.Markdown("*Upload one or more images for image-to-image generation. Leave empty for text-to-image mode.*")
+                
+                with gr.Accordion("⚙️ Advanced Settings", open=False):
+                    seed = gr.Slider(
+                        label="Seed",
+                        minimum=0,
+                        maximum=MAX_SEED,
+                        step=1,
+                        value=42,
+                    )
+                    
+                    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,
+                            info="Must be a multiple of 32"
+                        )
+                        
+                        height = gr.Slider(
+                            label="Height",
+                            minimum=256,
+                            maximum=MAX_IMAGE_SIZE,
+                            step=32,
+                            value=1024,
+                            info="Must be a multiple of 32"
+                        )
+                    
+                    with gr.Row():
+                        num_inference_steps = gr.Slider(
+                            label="Number of inference steps",
+                            minimum=1,
+                            maximum=100,
+                            step=1,
+                            value=50,
+                        )
+                        
+                        guidance_scale = gr.Slider(
+                            label="Guidance scale",
+                            minimum=0.0,
+                            maximum=10.0,
+                            step=0.1,
+                            value=1.5,
+                        )
+                
+            with gr.Column():
+                result = gr.Image(label="Result", show_label=False)
+        
+        gr.Markdown("### 📝 Text-to-Image Examples")
+        gr.Examples(
+            examples=examples_t2i,
+            fn=infer,
+            inputs=[prompt],
+            outputs=[result, seed],
+            cache_examples="lazy"
+        )
+        
+        gr.Markdown("### 🖼️ Image-to-Image Example Prompts")
+        gr.Markdown("*Upload an image above and use one of these prompts:*")
+        for example in examples_i2i:
+            gr.Markdown(f"- {example[0]}")
+
+    # Auto-update dimensions when images are uploaded
+    input_images.upload(
+        fn=update_dimensions_from_image,
+        inputs=[input_images],
+        outputs=[width, height]
+    )
+
+    gr.on(
+        triggers=[run_button.click, prompt.submit],
+        fn=infer,
+        inputs=[prompt, input_images, seed, randomize_seed, width, height, num_inference_steps, guidance_scale],
+        outputs=[result, seed]
+    )
+
+demo.launch(theme=gr.themes.Citrus(), css=css)