app.py

import gradio as gr
import torch
from diffusers import LTXPipeline
from diffusers.utils import export_to_video
import tempfile
import random

# Load the LTX Video model
pipe = LTXPipeline.from_pretrained("Lightricks/LTX-Video", torch_dtype=torch.bfloat16)
pipe.to("cuda")

def generate_video(prompt, negative_prompt, height, width, num_frames, num_inference_steps, seed):
    if seed == -1:
        seed = random.randint(0, 2**32 - 1)
    
    generator = torch.Generator(device="cuda").manual_seed(seed)

    video = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        height=height,
        width=width,
        num_frames=num_frames,
        num_inference_steps=num_inference_steps,
        generator=generator
    ).frames[0]

    # Export video to temporary file
    with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmpfile:
        export_to_video(video, tmpfile.name, fps=24)
        return tmpfile.name

# Gradio Interface
title = "LTX-Video Generator"
description = "Generate high-quality videos from text using the Lightricks LTX-Video model."

with gr.Blocks(title=title) as demo:
    gr.Markdown(f"## {title}\n{description}")
    
    with gr.Row():
        prompt = gr.Textbox(label="Prompt", value="A woman with long brown hair and light skin smiles at another woman...", lines=5)
        negative_prompt = gr.Textbox(label="Negative Prompt", value="worst quality, inconsistent motion, blurry, jittery, distorted", lines=5)

    with gr.Row():
        height = gr.Slider(minimum=64, maximum=720, step=32, value=480, label="Height")
        width = gr.Slider(minimum=64, maximum=1280, step=32, value=704, label="Width")
        num_frames = gr.Slider(minimum=9, maximum=257, step=8, value=161, label="Number of Frames")
        num_inference_steps = gr.Slider(minimum=10, maximum=100, step=1, value=50, label="Inference Steps")
        seed = gr.Number(value=-1, label="Seed (set -1 for random)")

    generate_btn = gr.Button("Generate Video")
    output_video = gr.Video(label="Generated Video")

    generate_btn.click(
        fn=generate_video,
        inputs=[prompt, negative_prompt, height, width, num_frames, num_inference_steps, seed],
        outputs=output_video
    )

demo.launch()