LTX 2.3 on Mac: M-Series, MacBook, Mac Mini & MLX

virtuavixen No Comments

Yes, you can run LTX 2.3 on a Mac — but with significant caveats. Apple Silicon (M-series chips) supports LTX 2.3 through MLX-based ports and Draw Things, but the experience is a step behind NVIDIA: slower generation, no audio support, and feature gaps on advanced workflows. This guide covers what works, what doesn't, and which Mac models are actually worth using.

If you have an Apple Silicon Mac and want to use LTX 2.3 today, the easiest path is our VirtuaVixen Studio — runs in Safari on Mac, M2 or later, full audio, full LoRA stack, no setup. 160 free tokens daily. The ComfyUI Workflow Pack can be used on Mac via Linux VMs but native ComfyUI on Apple Silicon is still rough. Discord for Mac-specific help.

What Works on Mac

Mac ModelRAMQuantTime per 5s clipAudio support
M4 Max (40-core GPU)64 GB+GGUF Q4~15–20 minLimited
M3 Max48 GB+GGUF Q4~20–30 minNo
M2 Pro / Max32 GB+GGUF Q4_0~30–45 minNo
M2 / M3 base16 GBNot recommendedNo
Mac Mini M424 GBGGUF Q4_0~25 minNo
Intel MacsNot supportedNo

Option 1: Draw Things

Draw Things is a free Mac app that supports LTX 2.3 through their built-in model browser. Easiest setup on Mac:

  1. Download Draw Things from the Mac App Store (free)
  2. Open the model browser, search “LTX 2.3”
  3. Download the model and any LoRAs you want
  4. Generate from the prompt panel

Limitations: the Draw Things port is the distilled model only (no Dev), no audio generation (silent video output), and the LoRA selection is limited to what they package. Reasonable for SFW work; less useful for NSFW where the LoRA stack matters.

Option 2: MLX Port

MLX is Apple's machine learning framework, comparable to PyTorch but native to Apple Silicon. Community ports of LTX 2.3 to MLX exist on GitHub, with names like mlx-ltx or similar. Setup is non-trivial — you need Xcode, Swift toolchain, and to compile some kernels.

Performance is faster than ComfyUI-on-Linux-VM running on the same Mac, but still 3–5× slower than a 4090 on bare-metal Linux. Audio support is not yet implemented — MLX ports are video-only.

Option 3: ComfyUI on Mac (via PyTorch + MPS)

ComfyUI runs on macOS with PyTorch's MPS (Metal Performance Shaders) backend. Setup similar to Linux:

git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
pip3 install -r requirements.txt
python3 main.py

Issues you'll hit:

  • Some custom nodes are CUDA-only and won't work on MPS — including parts of the LTX 2.3 audio pipeline
  • FP8 quants don't work on MPS — you need GGUF Q4 or Q8
  • Memory management is shakier; ~2× the system RAM as your effective VRAM is the rough rule
  • Custom triton kernels (used in some KJ nodes) aren't available on Mac — workflows that depend on them fail to load

For full LTX 2.3 functionality (audio, lipsync, advanced LoRAs), ComfyUI on Mac currently doesn't get you there. Use a Linux machine, a cloud GPU, or our hosted Studio.

Option 4: Linux VM (Asahi Linux or UTM)

Some users run ComfyUI inside a Linux VM on Mac (UTM with Ubuntu). The VM doesn't have GPU acceleration so this is CPU-only — completely unusable for real LTX 2.3 generation.

Asahi Linux (native Linux on Apple Silicon) has the same MPS/MLX limitations as macOS for AI workloads. Not a real solution.

The Honest Recommendation

If you have a Mac and want to use LTX 2.3 today:

  • For SFW use: Draw Things is the simplest path. Slow but works.
  • For NSFW use with audio: use our browser-based Studio. Runs perfectly on any Mac with Safari/Chrome, no install.
  • For programmatic use: rent a RunPod or Vast.ai instance from your Mac via SSH. See cloud GPU options.
  • For local sustained work: get a PC with NVIDIA GPU. The Mac path will frustrate you.

When the Mac Path Is Worth It

Three scenarios where running LTX 2.3 on Mac is genuinely useful:

  • You're occasionally generating SFW clips and don't want a separate machine
  • You're traveling with a MacBook Pro and don't want to depend on cloud connectivity
  • You're prototyping a Mac-native AI app and want LTX 2.3 inference for it

For everything else, NVIDIA on Linux/Windows is faster and more featureful, or our Studio is easier and free.

Try Studio Instead

Our VirtuaVixen Studio works perfectly on any Mac (M-series or Intel) — full LTX 2.3 NSFW workflow set, audio, lipsync, FML keyframes, all running on rented 48 GB GPUs in our datacenter. Open Safari, sign up free, generate. 160 daily tokens, no setup, no battery drain on your laptop.

If you do decide to set up locally on a Linux/Windows PC, the Workflow Pack ships an installer that handles all the model and LoRA downloads in one command.

Related Reading

Leave a comment

Are you 18 or older?

You must be 18 years or older to access this website.

👑 AI Studio ×

Categories