Deploy Gradio app with multiple files

app.py

@@ -0,0 +1,54 @@
+import gradio as gr
+from utils import generate_video
+import spaces
+import os
+from huggingface_hub import login
+
+# Login to Hugging Face (users need to provide their token)
+hf_token = os.getenv("HF_TOKEN")
+if hf_token:
+    login(hf_token)
+else:
+    raise ValueError("Please set HF_TOKEN environment variable for model access")
+
+# Compile the model with ZeroGPU AoT for speed
+from utils import compile_model
+compiled_model = compile_model()
+
+@spaces.GPU(duration=120)  # Allow time for video generation
+def generate(prompt: str, user: str):
+    # Generate video from prompt
+    video_path = generate_video(prompt, compiled_model)
+    # Save to user's account (placeholder - integrate with HF Hub or database)
+    return video_path
+
+with gr.Blocks(title="Fast Text-to-Video Generator", theme=gr.themes.Soft()) as demo:
+    gr.HTML("""
+    <h1 style='text-align: center'>Fast Text-to-Video Generator</h1>
+    <p style='text-align: center'>Login to your account to generate personalized videos. Built with <a href='https://huggingface.co/spaces/akhaliq/anycoder' target='_blank'>anycoder</a>.</p>
+    """)
+    
+    login_btn = gr.LoginButton("Sign in with Hugging Face")
+    
+    with gr.Row(visible=False) as main_ui:
+        prompt_input = gr.Textbox(label="Enter your text prompt", placeholder="A cat playing in a garden")
+        generate_btn = gr.Button("Generate Video")
+        video_output = gr.Video(label="Generated Video")
+        status = gr.Textbox(label="Status", interactive=False)
+    
+    def show_ui():
+        return gr.Row(visible=True)
+    
+    def create_video(prompt, request: gr.Request):
+        if not request.username:
+            return None, "Please login first"
+        status = "Generating video... (this is very fast!)"
+        video = generate(prompt, request.username)
+        status = "Video generated successfully!"
+        return video, status
+    
+    login_btn.click(show_ui, outputs=main_ui)
+    generate_btn.click(create_video, inputs=[prompt_input], outputs=[video_output, status])
+
+if __name__ == "__main__":
+    demo.launch()

config.py

@@ -0,0 +1,2 @@
+MODEL_ID = "stabilityai/stable-video-diffusion-img2vid-xt"
+HF_TOKEN_ENV = "HF_TOKEN"

requirements.txt

@@ -0,0 +1,8 @@
+gradio
+torch
+diffusers
+accelerate
+transformers
+huggingface-hub
+imageio
+spaces

utils.py

@@ -0,0 +1,43 @@
+import torch
+from diffusers import StableVideoDiffusionPipeline
+from diffusers.utils import load_image
+import spaces
+
+def compile_model():
+    # Load the model
+    model_id = "stabilityai/stable-video-diffusion-img2vid-xt"
+    pipe = StableVideoDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16, variant="fp16")
+    pipe.to('cuda')
+    
+    @spaces.GPU(duration=1500)  # AoT compilation
+    def compile_transformer():
+        # Capture example inputs
+        with spaces.aoti_capture(pipe.unet) as call:
+            image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/rocket.png")
+            pipe(image).frames
+        
+        # Export and compile
+        exported = torch.export.export(
+            pipe.unet,
+            args=call.args,
+            kwargs=call.kwargs,
+        )
+        return spaces.aoti_compile(exported)
+    
+    compiled_unet = compile_transformer()
+    spaces.aoti_apply(compiled_unet, pipe.unet)
+    return pipe
+
+def generate_video(prompt: str, pipe):
+    # For simplicity, use a placeholder image; in real app, generate image from text first
+    image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/rocket.png")  # Placeholder
+    
+    # Generate video
+    frames = pipe(image, decode_chunk_size=8).frames[0]
+    
+    # Save as video (placeholder path)
+    import imageio
+    video_path = f"/tmp/generated_video_{hash(prompt)}.mp4"
+    imageio.mimsave(video_path, frames, fps=7)
+    
+    return video_path