app.py

import gradio as gr
import torch
import spaces
from diffusers import LTXConditionPipeline
from diffusers.utils import export_to_video
from gtts import gTTS
from pydub import AudioSegment
import whisper
import ffmpeg
import os

# Load pipeline
pipe = LTXConditionPipeline.from_pretrained(
    "Lightricks/LTX-Video-0.9.7-distilled", torch_dtype=torch.float16
)
pipe.to("cuda")

@spaces.GPU(duration=120)
def generate_video(prompt):
    generator = torch.Generator("cuda").manual_seed(42)

    # Generate latent video
    latents = pipe(
        prompt=prompt,
        width=512,
        height=512,
        num_frames=24,
        output_type="latent",
        generator=generator,
        num_inference_steps=7
    ).frames

    # Decode frames
    frames = pipe(
        prompt=prompt,
        latents=latents,
        num_frames=24,
        output_type="pil",
        generator=generator,
        num_inference_steps=7
    ).frames[0]

    # Save as video
    video_path = "output.mp4"
    export_to_video(frames, video_path, fps=12)

    # TTS
    tts = gTTS(text=prompt, lang='en')
    tts.save("voice.mp3")
    AudioSegment.from_mp3("voice.mp3").export("voice.wav", format="wav")

    # Subtitles
    model = whisper.load_model("base")
    result = model.transcribe("voice.wav", language="en")
    with open("subtitles.srt", "w") as f:
        f.write(result["srt"])

    # Merge audio + subtitles into video
    ffmpeg.input(video_path).output(
        "final.mp4",
        vf="subtitles=subtitles.srt",
        i="voice.mp3",
        c="copy",
        shortest=None,
        loglevel="error"
    ).run()

    return "final.mp4"

# Gradio UI
demo = gr.Interface(fn=generate_video, inputs="text", outputs=gr.Video())
demo.launch()