import os
import torch
import gradio as gr
from diffusers import DiffusionPipeline
import soundfile as sf

# Читаем токен из секрета Hugging Face
HF_HUB_TOKEN = os.environ["HF_HUB_TOKEN"]

# Загружаем pipeline из подпапки GFengG
pipe = DiffusionPipeline.from_pretrained(
    "MeiGen-AI/MeiGen-MultiTalk",
    subfolder="GFengG",
    token=HF_HUB_TOKEN,
    torch_dtype=torch.float16
).to("cuda" if torch.cuda.is_available() else "cpu")

def generate(prompt: str):
    out = pipe(prompt)       # см. README: может быть text2speech, audio2audio и т.п.
    audio = out.audios[0]
    sf.write("generated.wav", audio, samplerate=pipe.unet.config.sample_rate)
    return "generated.wav"

iface = gr.Interface(
    fn=generate,
    inputs=gr.Textbox(lines=2, placeholder="Введите текст..."),
    outputs=gr.Audio(source="file", type="filepath"),
    title="MeiGen-MultiTalk Demo"
)

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