app.py

# app.py
import os, sys
import gradio as gr
import torch

os.environ.setdefault("HF_HUB_ENABLE_HF_TRANSFER", "1")
MODEL_REPO = "vibevoice/VibeVoice-1.5B"

demo_app = None
model_ready = False

def load_real_demo():
    global demo_app, model_ready
    if not model_ready:
        from demo import gradio_demo
        # 假设 gradio_demo.py 里有 main(return_app=True) 可以返回一个 Blocks
        demo_app = gradio_demo.main(model_path=MODEL_REPO, return_app=True)
        model_ready = True
    return demo_app

def launch_wrapper():
    load_real_demo()
    return demo_app

with gr.Blocks() as app:
    gr.Markdown("## 🎙️ VibeVoice Demo (lazy-load)")
    btn = gr.Button("🚀 Load full demo")
    out = gr.HTML()
    btn.click(fn=lambda: "<b>Reload page to see full demo</b>", outputs=out)

if __name__ == "__main__":
    app.queue().launch(server_name="0.0.0.0", server_port=7860)