README.md

metadata

license: other
license_name: ltx-video-license
license_link: https://huggingface.co/Lightricks/LTX-2/blob/main/LICENSE.md
library_name: mlx
tags:
  - video-generation
  - mlx
  - apple-silicon
  - ltx-2
  - distilled
  - quantized

LTX-2 19B Distilled (4-bit) - MLX

This is a 4-bit quantized version of the LTX-2 19B Distilled model, optimized for Apple Silicon using MLX.

Model Description

LTX-2 is a state-of-the-art video generation model from Lightricks. This version has been quantized to 4-bit precision for efficient inference on Apple Silicon devices with MLX.

Key Features

• Pipeline: Distilled (faster generation, fewer steps required)
• Quantization: 4-bit precision
• Framework: MLX (Apple Silicon optimized)
• Memory: ~19GB VRAM required

Usage

Installation

pip install git+https://github.com/CharafChnioune/mlx-video.git

Command Line

# Basic generation
mlx-video --prompt "A beautiful sunset over the ocean" \
    --model-repo AITRADER/ltx2-distilled-4bit-mlx \
    --pipeline distilled \
    --height 512 --width 512 \
    --num-frames 33

Python API

from mlx_video import generate_video

video = generate_video(
    prompt="A beautiful sunset over the ocean",
    model_repo="AITRADER/ltx2-distilled-4bit-mlx",
    pipeline="distilled",
    height=512,
    width=512,
    num_frames=33,
)

Model Files

• ltx-2-19b-distilled-mlx.safetensors - Main model weights (4-bit quantized)
• quantization.json - Quantization configuration
• config.json - Model configuration
• layer_report.json - Layer information

Performance

Resolution	Frames	Time
512x512	33	~30s on M3 Max
768x512	33	~45s on M3 Max

License

This model is released under the LTX Video License.

Acknowledgements

• Lightricks for the original LTX-2 model
• MLX team at Apple for the framework
• mlx-video for the MLX conversion