Claude Code Club
Video walkthroughs

LTX-2.5: The Free Open AI Video Model You Can Run Yourself

6 minute readUpdated October 2026Explore more

TL;DR

LTX-2.5 is an open-weights AI video model from Lightricks. It makes the video and the sound in one pass, you can download the weights for free from Hugging Face (you need a free Hugging Face account and must accept the model terms first), and you can train it on your own style with the LoRA trainer in the official GitHub repo. It needs a strong NVIDIA graphics card (about 32GB of video memory) or a rented cloud GPU. Once it runs, every extra take costs you nothing but electricity or GPU time, with no per-video credits.

Learn Claude Code. Earn income. Only $9.

👉 https://www.skool.com/claudecodeclub

What LTX-2.5 is

LTX-2.5 is the newest open video model from Lightricks. You describe a shot, or give it an image, and it generates the video and the matching sound together: motion, dialogue, ambience and music in one pass. The weights are public, so you can run it on your own machine instead of paying per clip on a hosted tool like Veo or Kling.

Lightricks/LTX-2

9,568 stars on October 2, 2026. The official code: inference pipelines plus the LoRA trainer.

Lightricks/LTX-2.5 on Hugging Face

The model weights. Over 1.5 million downloads on October 2, 2026. Free, but gated: log in with a free Hugging Face account and accept the terms on the model page before you can download.

What you need before you start

  • A strong NVIDIA graphics card. The ComfyUI guide asks for a CUDA GPU with 32GB+ of video memory. Lower-memory cards can run it with FP8 and offloading, slower.
  • Disk space. The main weights are about 66GB, and the ComfyUI guide recommends 100GB+ free for models and cache.
  • No strong GPU? Rent one. Cloud GPU services like Runpod let you run LTX-2.5 by the hour. Runpod has its own LTX-2.5 setup guide.
  • A Mac is not the right machine for this. The fast paths are built for NVIDIA CUDA.

Easiest way to run it: ComfyUI

LTX-2.5 had day-one ComfyUI support. If you already use ComfyUI, this is the shortest path:

  1. 1Open ComfyUI and click Manager (or press Ctrl+M).
  2. 2Install the official LTX-2 workflows that Lightricks publishes in Comfy Manager.
  3. 3Open an LTX-2.5 template, type the shot you want, and run it. The first run downloads the model files.
Lightricks/ComfyUI-LTXVideo

Official ComfyUI nodes and example workflows for LTX-2.5, including low-VRAM loaders.

Run it from the official repo

Or let Claude Code set it up from the README on a Linux machine or rented GPU:

promptSet up LTX-2.5 for me by following the Quick Start in this README: https://github.com/Lightricks/LTX-2 . Check my GPU memory and free disk first, use the DistilledPipeline, and use fp8 if my GPU has less than 32GB.

The README's own steps, if you'd rather run them yourself:

bashgit clone https://github.com/Lightricks/LTX-2.git
cd LTX-2
uv sync --extra natten
hf auth login
hf download Lightricks/LTX-2.5 --local-dir models/ltx-2.5

Then run `uv run python -m ltx_pipelines.distilled` with the model paths and your prompt (the full command is in the README). If Hugging Face returns 401 or 403, accept the model terms on the model page and log in with a Read token.

How to describe the shot

Lightricks' prompting advice: write one flowing paragraph, under 200 words, like a cinematographer's shot list. Start with the main action, then movement, what things look like, the background, the camera angle and move, and the lighting. Describe sounds and dialogue too, since the audio is generated with the video.

promptWrite me an LTX-2.5 prompt for [your shot]. One paragraph, under 200 words: main action first, then movement, appearance, background, camera angle and movement, lighting and colors, and the sounds we should hear.

Train it on your own style

The repo includes ltx-trainer for LoRA, full fine-tuning and IC-LoRA. Its README recommends Linux with CUDA and an 80GB GPU for the standard config, or a low-VRAM config for 32GB cards like the RTX 5090. In practice, that means a rented GPU for most people.

Quick start

  1. 1Check your GPU memory and free disk, or pick a cloud GPU.
  2. 2Install ComfyUI and the official LTX-2 workflows, or follow the GitHub Quick Start.
  3. 3Describe one shot in a single paragraph and generate it.
  4. 4Make as many takes as you like, then keep the best one.
Learn Claude Code. Earn income. Only $9.

👉 https://www.skool.com/claudecodeclub

Common questions

  • Is LTX-2.5 really free?

    The weights are free to download and use under the LTX-2.x community license for individuals and companies under $10 million in yearly revenue. You pay for your own hardware or rented GPU time, not per video.

  • Can I run LTX-2.5 on a Mac or a normal laptop?

    Not well. It is built for NVIDIA GPUs with about 32GB of video memory. On a Mac or a normal laptop, rent a cloud GPU instead.

  • Does LTX-2.5 make sound too?

    Yes. It generates the video and the audio together in one pass, including ambience, music and dialogue you describe in the prompt.

  • Where is the code and where are the weights?

    The code and LoRA trainer are on GitHub at Lightricks/LTX-2. The weights are on Hugging Face at Lightricks/LTX-2.5.

Keep going

Get more free AI tools like this inside Claude Code Club for $9/mo

Get 650+ plug-and-play skills, MCPs & prompts, plus 8,000+ members - $9/mo, cancel anytime.

Join the Club