LTX Video Generator

Local text-to-video and image-to-video on Apple Silicon. LTX-2 / 2.3, LTX-2.5, and MiniMax H3.

Download v2.3.69 All Releases View on GitHub

What’s New in 2.3.69

LTX-2.5 — Distilled (ltx25_distilled, ~100GB), Dev two-stage (ltx25_dev, same pack), and Distilled Q8 DiT (ltx25_distilled_ditq8). Gemma 4 is bundled. The app installs dgrauet/ltx-2-mlx 0.15+ only when you pick a 2.5 model.
MiniMax H3 — Native h3.c (minimax_h3 BF16 ~144GB, minimax_h3_int8 ~92GB, minimax_h3_turbo folded 6-step). License dialog on first select/generate; weights auto-download into the Hugging Face cache. Build ./h3 yourself (not in the DMG).
Unchanged 2.3 path — LTX-2 / 2.3 stay on mlx-video-with-audio. REST POST /generate accepts the new model_id values.

Release notes · Installation · Architecture

Native macOS Experience

SwiftUI app. Generation is a local subprocess: mlx-video-with-audio for LTX-2 / 2.3, ltx-2-mlx for 2.5, ./h3 for MiniMax H3.

Key Features

Apple Silicon Native - MLX for LTX; Metal via h3.c for H3
LTX-2, LTX-2.3, LTX-2.5, MiniMax H3 - 2.3 Q4 default (~22GB) except ≤16GB Macs (12GB pack); 2.5 / 12GB via ltx-2-mlx; H3 via native h3.c
Text-to-Video - Generate videos from text descriptions
Image-to-Video - Animate images; first/last frame and multi-image keyframes on the LTX path
Local REST API - 127.0.0.1:8420 (ltx25_distilled, ltx25_dev, ltx25_distilled_ditq8, minimax_h3, minimax_h3_int8, minimax_h3_turbo)
Gemma Prompt Enhancement - Optional AI rewrites prompts for better results; uncensored enhancer avoids content filters
Voiceover Narration - Add TTS audio using ElevenLabs (cloud) or MLX-Audio (local)
Background Music - 54 genre presets for AI-generated instrumental music via ElevenLabs
Auto Package Installer - Missing Python packages detected and installed with one click
Generation Queue - Queue multiple videos and track progress in real-time
Smart History - Browse, preview, and manage all your generated videos
Flexible Presets - Quick access to common configurations or customize every parameter

Quick Start

Download the app from the Releases page
Open Preferences and click Auto Detect to find Python
Install packages if prompted (one-click install available)
Generate your first video! (model downloads on first run)

System Requirements

Requirement	Minimum	Recommended
macOS	14.0+	15.0+
Processor	Apple M1	Apple M2 Pro/M3/M4
Unified Memory	32GB (2.3 Q4)	64GB+ (2.5 / H3)
Storage	40GB free (2.3 Q4)	150GB+ (2.5 ~100GB or H3 ~144GB)
Python	3.10+	3.12+

First run: the selected model downloads from Hugging Face on first generate (default 2.3 Distilled Q4 is ~22GB). LTX-2.5 is ~100GB; MiniMax H3 is ~144GB and requires accepting the MiniMax H3 Community License. Cache is ~/.cache/huggingface/ unless you set a Model Cache Directory.

Sample Results

Generate videos like:

“A river flowing through a misty forest at dawn”
“The camera slowly pans across a futuristic cityscape”
“Golden leaves falling in slow motion against a blue sky”

Getting Help

Installation Guide - Complete setup instructions
Usage Guide - Learn how to get the best results (includes LTX-2.5 prompting)
Example Prompts - Copy-paste LTX-2.5-style prompts
Parameters Reference - Understand all settings
Troubleshooting - Common issues and solutions
Architecture - Technical details of the pipeline and models

Contributing

LTX Video Generator is open source! Contributions, issues, and feature requests are welcome on GitHub.