app.py

import os
import streamlit as st
import subprocess
from gtts import gTTS
import cv2
from huggingface_hub import InferenceClient
import torch

device = 'cuda' if torch.cuda.is_available() else 'cpu'
print('Using {} for inference.'.format(device))

client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")

def generate_output(prompt):
    if not prompt:
        return None, "El campo de la pregunta es obligatorio."

    response = client.text_generation(prompt, max_new_tokens=50, temperature=0.6)
    gpt3_output = response.strip()
    personalized_response = f"{gpt3_output}"
    
    try:
        tts = gTTS(personalized_response, lang='es')
        audio_path = "audio.mp3"
        tts.save(audio_path)
    except Exception as e:
        return None, f"No se pudo generar el audio: {str(e)}"

    video_path = "video.mp4"
    command = f"CUDA_VISIBLE_DEVICES='' python3 inference.py --checkpoint_path checkpoints/wav2lip_gan.pth --face face.jpg --audio {audio_path} --outfile {video_path} --nosmooth --resize_factor 4"
    process = subprocess.run(command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
    
    if process.returncode != 0:
        error_message = process.stderr
        return None, f"No se pudo generar el video: {error_message}"

    if os.path.isfile(video_path):
        return video_path, None
    return None, "No se pudo generar el video"

st.title("Lypsinc + Inteligencia Artificial")
prompt = st.text_input("Pregunta")

if st.button("Generar Video"):
    video_path, error_message = generate_output(prompt)
    if error_message:
        st.error(f"Error: {error_message}")
    else:
        with open(video_path, "rb") as video_file:
            st.video(video_file.read())