app.py

import streamlit as st
from moviepy.editor import *
from transformers import pipeline
from TTS.api import TTS
import tempfile, os

st.title("📝 Text-to-Video App with Voice Clone")

# Caching for faster reloads
@st.cache_resource()
def load_models():
    video_gen = pipeline('text-to-video-generation', model='cerspense/zeroscope_v2_XS')  # extra small version
    tts_model = TTS("tts_models/en/ljspeech/tacotron2-DDC", progress_bar=False)  # lighter TTS model
    return video_gen, tts_model

video_gen, tts_model = load_models()

# Input
input_text = st.text_area("Enter short text (max 100 chars):", max_chars=100)
voice_file = st.file_uploader("Upload your voice sample (short WAV):", type=["wav"])

if st.button("Generate"):
    if input_text and voice_file:
        with st.spinner("Creating video (may take a minute)..."):
            # Short video (15 frames only)
            video_output = video_gen(input_text, num_frames=15)
            video_tensor = video_output["video"]
            video_np = (video_tensor * 255).astype('uint8')

            video_filename = tempfile.mktemp(suffix=".mp4")
            clips = [ImageClip(frame).set_duration(0.2) for frame in video_np]
            video_clip = concatenate_videoclips(clips)
            video_clip.write_videofile(video_filename, fps=5)

            # Short audio clip
            voice_path = tempfile.mktemp(suffix=".wav")
            audio_filename = tempfile.mktemp(suffix=".wav")
            with open(voice_path, "wb") as f:
                f.write(voice_file.read())

            tts_model.tts_to_file(text=input_text, speaker_wav=voice_path, file_path=audio_filename)

            # Combine video and audio
            final_clip = VideoFileClip(video_filename).set_audio(AudioFileClip(audio_filename))
            final_video_path = tempfile.mktemp(suffix=".mp4")
            final_clip.write_videofile(final_video_path, fps=5)

            st.video(final_video_path)

            # Cleanup
            for f in [video_filename, audio_filename, voice_path, final_video_path]:
                os.remove(f)
    else:
        st.warning("Provide both text and a voice sample.")