from huggingface_hub import snapshot_download
import gradio as gr
import subprocess
import os
import uuid

def setup_models():
    if not os.path.exists("checkpoints"):
        print("Downloading model...")
        snapshot_download(repo_id="deepinsight/first-order-model", local_dir="checkpoints")

setup_models()

def generate(text, image):
    session = str(uuid.uuid4())[:8]
    os.makedirs(f"results/{session}", exist_ok=True)
    image_path = f"results/{session}/avatar.jpg"
    image.save(image_path)
    audio_path = f"results/{session}/audio.wav"
    tts_cmd = f'tts --text "{text}" --out_path {audio_path}'
    subprocess.run(tts_cmd, shell=True, check=True)
    video_cmd = f'python checkpoints/inference.py --driven_audio {audio_path} --source_image {image_path} --result_dir results/{session}'
    subprocess.run(video_cmd, shell=True, check=True)
    return f"results/{session}/video.mp4"

gr.Interface(
    fn=generate,
    inputs=[gr.Textbox(label="Script"), gr.Image(label="Avatar Image", type="pil")],
    outputs=gr.Video(label="Generated Video"),
    title="Faceless Video Generator (Free)",
    description="Type your message + upload an image. Get a talking avatar video!"
).launch()