Installation Guide

Complete setup instructions for LTX Video Generator.

Table of contents

  1. Download the App
  2. Configure Python
    1. Step 1: Open Preferences
    2. Step 2: Validate Setup
    3. Step 3: Install Missing Packages
  3. LTX-2.5 and LTX-2.3 12GB (ltx-2-mlx)
  4. MiniMax H3 (h3.c)
  5. First Run - Model Download
    1. What to Expect
    2. Download Progress
    3. Storage Location
  6. Optional: Gemma Prompt Enhancement
  7. Verify Installation
  8. Installing Python
    1. Homebrew (Recommended)
    2. pyenv
    3. System Python
  9. Troubleshooting
    1. “Python not found”
    2. “Missing packages” after install
    3. “Out of memory” during generation

Download the App

  1. Go to the Releases page
  2. Download the latest .dmg file
  3. Open the DMG and drag LTX Video Generator to your Applications folder
  4. Right-click the app and select Open (required for first launch of notarized apps)

Configure Python

LTX Video Generator needs Python 3.10+ with MLX and related packages.

Step 1: Open Preferences

  1. Launch LTX Video Generator
  2. Open Preferences (⌘,)
  3. Click Auto Detect to find your Python installation

The app will search common locations including Homebrew, pyenv, conda, and system Python.

Step 2: Validate Setup

Click Validate Setup to check for required packages:

  • mlx - Apple’s machine learning framework
  • mlx-vlm - Vision-language models for MLX
  • mlx-video-with-audio - Unified audio-video generation (LTX-2 / 2.3)
  • ltx-2-mlx - LTX-2.5 only (git install, not required for 2.3)
  • transformers - Hugging Face transformers
  • safetensors - Fast tensor serialization
  • huggingface_hub - Model downloading
  • numpy - Numerical computing
  • opencv-python - Video encoding
  • tqdm - Progress bars

Step 3: Install Missing Packages

If packages are missing, you have two options:

Option A: One-Click Install (Recommended)

Click the Install Missing Packages button in Preferences. The app will run pip install automatically.

Option B: Manual Install

pip install mlx mlx-vlm mlx-video-with-audio transformers safetensors huggingface_hub numpy opencv-python tqdm

If using a virtual environment, make sure to activate it first, or point the app to the venv’s Python executable.

LTX-2.5 and LTX-2.3 12GB (ltx-2-mlx)

ltx-2-mlx is opt-in and is not installed for mlx-video-with-audio-only users. When you select LTX-2.5 or LTX-2.3 12GB RAM Optimized, the app tries to pip-install from git into the venv:

pip install \
  "git+https://github.com/dgrauet/ltx-2-mlx.git@v0.15.2#subdirectory=packages/ltx-core-mlx" \
  "git+https://github.com/dgrauet/ltx-2-mlx.git@v0.15.2#subdirectory=packages/ltx-pipelines-mlx" \
  "mlx-lm>=0.31.2"

Or clone to ~/projects/ltx-2-mlx, run uv sync --all-extras, and enable Use local ltx-2-mlx repo in Preferences.

MiniMax H3 (h3.c)

H3 does not use a Python inference runtime. First Generate clones antirez/h3.c (BF16 / Turbo) or Shedrackeze002/h3.c-int8 (int8, pin 200eed4) into Application Support and runs make (needs Xcode Command Line Tools). h3 binary path / H3 model directory override BF16 only. Selecting H3 shows an Accept / Read license dialog for the MiniMax H3 Community License (also covers the Turbo LoRA as a derivative).

Variant First-generate download
BF16 MiniMaxAI/MiniMax-H3 (~144GB)
int8 Comfy-Org int8 DiT (~20GB) + MiniMaxAI text encoder/VAEs (~72GB). Reuses a local BF16 snapshot for TE/VAE when present.
Turbo Official BF16 if missing, then larryvrh/MiniMax-H3-Turbo-Lora and an offline fold (~2–3GB APFS CoW)

If the MiniMaxAI repo is gated, run hf auth login first. US/EU/UK/KR users may need territory authorization.

First Run - Model Download

Important: On your first generation, the app downloads the model selected in Preferences from Hugging Face. This is a one-time download.

What to Expect

  1. Start a generation with any prompt
  2. Progress shows “Downloading model…” with percentage
  3. Download takes 30-60 minutes depending on connection speed
  4. Model is cached in ~/.cache/huggingface/hub/
  5. Subsequent runs skip the download

Download Progress

The app shows real-time download progress:

  • Downloading: 8.4GB / 42GB (20%)
  • Downloading: 4.0GB / 19.4GB (21%)

If download is interrupted, it will resume from where it left off.

Storage Location

Models are cached by Hugging Face in folders such as:

~/.cache/huggingface/hub/models--notapalindrome--ltx2-mlx-av/
~/.cache/huggingface/hub/models--dgrauet--ltx-2.3-mlx-distilled-q4/
~/.cache/huggingface/hub/models--baa-ai--LTX-2.3-22B-RAM-12GB-MLX/
~/.cache/huggingface/hub/models--mlx-community--ltx-2.5-mlx/
~/.cache/huggingface/hub/models--MiniMaxAI--MiniMax-H3/
~/.cache/huggingface/hub/models--Comfy-Org--MiniMax-H3/
~/.cache/huggingface/hub/models--larryvrh--MiniMax-H3-Turbo-Lora/

To store new downloads on another disk:

  1. Open Preferences > General > Storage
  2. Set Model Cache Directory to a folder on the mounted disk
  3. Start generation after confirming the disk is still mounted

The app passes the selected directory to Hugging Face through HF_HOME and stores Hub downloads in its hub subfolder. Existing cache files are not moved automatically. Copy them from ~/.cache/huggingface/ into the selected directory before removing the originals.

If the selected folder is unavailable or not writable, generation stops with an error instead of silently downloading models to the internal disk.

To free up space later, you can delete this folder (the model will re-download on next use).

Optional: Gemma Prompt Enhancement

For better results, enable Settings > Generation > Enable Gemma Prompt Enhancement. Gemma rewrites your prompts with vivid details before generation.

If prompts with certain words (e.g. medical terms) return empty enhancement, enable Use uncensored enhancer. First run downloads ~7GB (TheCluster/amoral-gemma-3-12B-v2-mlx-4bit).

Verify Installation

To verify everything is working:

  1. Enter a simple prompt: "A river flowing through a forest"
  2. Use the Quick Preview preset (512x320, 49 frames)
  3. Click Generate
  4. Watch progress in the Queue sidebar
  5. Your video should appear when complete

Installing Python

If you don’t have Python installed, here are your options:

brew install python@3.12

Python will be at /opt/homebrew/bin/python3.12

pyenv 

# Install pyenv
brew install pyenv

# Install Python
pyenv install 3.12

# Set as default
pyenv global 3.12

System Python

macOS includes Python, but you may need to install Xcode Command Line Tools:

xcode-select --install

Troubleshooting

“Python not found”

  • Click Auto Detect in Preferences
  • Or manually enter the path to your Python executable
  • Verify it exists: which python3

“Missing packages” after install

  • Make sure you’re using the same Python the app is configured to use
  • Try: /path/to/your/python3 -m pip install mlx mlx-vlm mlx-video-with-audio transformers safetensors huggingface_hub numpy opencv-python tqdm

“Out of memory” during generation

  • Use smaller resolution (512x320)
  • Reduce frame count
  • Close other applications
  • 32GB RAM minimum required

See the Troubleshooting Guide for more solutions.

Back to top

Copyright © 2026 James Campbell. Distributed under the MIT License.

This site uses Just the Docs, a documentation theme for Jekyll.