# LTX-2 (LTX-2.5) — card for AI agents

Lightricks' open-weight model that generates video with sound from text, images or audio; runs locally, e.g. in ComfyUI.

Category: AI video generators and platforms · maker Lightricks · checked 2026-11-10

## Price
- Price: Free
- Free tier: Open weights, free to download after accepting the terms on Hugging Face

## License
LTX-2.x Community License Agreement — https://github.com/Lightricks/LTX-2/blob/main/LICENSE-2_x

- commercial: **conditions** (Commercial use with conditions)
  > Entities with annual revenues of at least $10,000,000 (the "Commercial Entities") are required to obtain a paid license
  (https://github.com/Lightricks/LTX-2/blob/main/LICENSE-2_x)
- attribution: **no** (No credit needed)
  > Licensor claims no rights in the Output you generate using LTX-2.x
  (https://github.com/Lightricks/LTX-2/blob/main/LICENSE-2_x)
- in app: **conditions** (Ship inside your app with conditions)
  > You may host for third parties remote access purposes (e.g. software-as-a-service), reproduce and distribute copies of LTX-2.x
  (https://github.com/Lightricks/LTX-2/blob/main/LICENSE-2_x)
- youtube: **conditions** (Monetized YouTube with conditions)
  > Entities with annual revenues of at least $10,000,000 (the "Commercial Entities") are required to obtain a paid license for any use
  (https://github.com/Lightricks/LTX-2/blob/main/LICENSE-2_x)

## Rules to follow
- Companies with annual revenues of $10M or more need a paid commercial license from Lightricks (licensing@ltx.io) for any non-research use.
- Do not remove or disable the watermarking, disclosures or content-provenance features; follow AI transparency laws such as the EU AI Act.
- Fine-tuned weights and other derivatives may only be shared under the same license, with a full copy of it.
- Accept the terms on the Hugging Face model page before downloading the weights.

## How to connect
- Download only
- Repository: https://github.com/Lightricks/LTX-2
- Website: https://huggingface.co/Lightricks/LTX-2.5

## Notes
Added by hand on Oct 25, 2026 (the asset robot had wrongly taken every Hugging Face model as already listed). Hugging Face API: 2,107,108 downloads and 9,277 likes; GitHub Lightricks/LTX-2: 9,873 stars.

## Numbers (AgentGid data)
- Gid Score: 34
- GitHub stars: 10,305

## Lines to paste into your prompt or AGENTS.md

```
- LTX-2 (LTX-2.5) (AI video generators and platforms). Lightricks' open-weight model that generates video with sound from text, images or audio; runs locally, e.g. in ComfyUI. Price: free; commercial use with conditions; no credit needed; ship inside your app with conditions.
  Connect: download from https://huggingface.co/Lightricks/LTX-2.5 (no API).
  Rule: Companies with annual revenues of $10M or more need a paid commercial license from Lightricks (licensing@ltx.io) for any non-research use.
  Rule: Do not remove or disable the watermarking, disclosures or content-provenance features; follow AI transparency laws such as the EU AI Act.
  Rule: Fine-tuned weights and other derivatives may only be shared under the same license, with a full copy of it.
  Rule: Accept the terms on the Hugging Face model page before downloading the weights.
  License: https://github.com/Lightricks/LTX-2/blob/main/LICENSE-2_x (checked 2026-11-10; details: https://agentgid.com/assets/ltx/)
```

## Links
- AgentGid page: https://agentgid.com/assets/ltx/
- Is it free for commercial use? https://agentgid.com/assets/ltx/commercial-use/
- Source: https://huggingface.co/Lightricks/LTX-2.5
- Source: https://github.com/Lightricks/LTX-2/blob/main/LICENSE-2_x
- Source: https://github.com/Lightricks/LTX-2
- Source: https://ltx.io


## LTX-2.5: video with sound from text or images

LTX-2.5 is a free-to-download AI model from Lightricks that makes short videos with matching sound from a text description, a picture or an audio track. It runs on your own powerful graphics card or inside ComfyUI. Companies earning $10 million a year or more need a paid license.

### Where to use it
- Talking clip from a written scene: Describe a shot, what people say and the sounds, and get a short clip where voice and picture match, for an ad draft or a social post.
- Bring a product photo to life: Start from a still picture, such as a product shot or an illustration, and turn it into a short moving clip.
- Video that follows your music or voice: Give it an audio file, like a song or a voice-over, and get a video made to go with that sound.
- Redo one part of a clip: Keep a clip you mostly like and regenerate only a few seconds of it, instead of starting over.
- Change what a speaker says: Rephrase a line in an existing clip while the speaker keeps the same voice and the lips match the new words.
- Drag-and-drop workflows in ComfyUI: Build video workflows by connecting boxes in ComfyUI instead of running commands, using the official LTX nodes.

### How to start
- Get the official code and install its parts with uv.

```
git clone https://github.com/Lightricks/LTX-2.git
cd LTX-2
uv sync --extra natten
```

- Accept the model terms on Hugging Face, sign in there with a read token and download the model files (about 66 GiB).
- Run the fast pipeline with a detailed prompt; on smaller cards add the memory-saving flags.

```
--quantization fp8-cast --offload cpu
```


### License
Free for people and companies under $10 million yearly revenue, for any purpose including paid work. Bigger companies need a paid license except for testing, and nobody may strip the AI-made labels or watermarks.

### Good to know
- It needs a strong graphics card and a lot of disk: the basic download alone is about 66 GiB.
- The download is gated; you need a free Hugging Face account, accepted terms and a read token.
- The code is open source, but the model files have their own license, which applies whenever you use them.
- Keep watermarks, provenance data and AI disclosures in your app and outputs; removing them can end your license.
- Posting generated content in places like an automated account needs a clear note that a machine made it.

Checked 2026-11-10 against: https://huggingface.co/Lightricks/LTX-2.5, https://huggingface.co/api/models/Lightricks/LTX-2.5, https://raw.githubusercontent.com/Lightricks/LTX-2/main/README.md, https://raw.githubusercontent.com/Lightricks/LTX-2/main/LICENSE-2_x, https://raw.githubusercontent.com/Lightricks/LTX-2/main/packages/ltx-pipelines/docs/installation.md, https://api.github.com/repos/Lightricks/LTX-2, https://ltx.io

### Mistakes to avoid
- Don't: Downloading Google's stock Gemma 4 as the text encoder → Do: Use the bundled gemma4-12b-with-proj-ltx-2.5-bf16.safetensors from the LTX-2.5 repository (Loading checks the encoder version, so the stock release is not a substitute.)
- Don't: Passing the full dev transformer to the DFR pipeline → Do: Use the distilled transformer plus the detailing IC-LoRA for DFR (The README says DFR uses the same distilled transformer and warns against the dev one.)
- Don't: --width 3840 --height 2160 for 4K → Do: --width 3840 --height 2176 (The README gives 2176, not 2160, as the 4K height.)
- Don't: Mixing LTX-2.3 checkpoints or LoRAs with LTX-2.5 files → Do: Keep each model's files and LoRAs together (Files are not interchangeable between the two models.)
- Don't: Calling a pipeline from Python without torch.inference_mode() → Do: Wrap the whole call, including prompt encoding, in torch.inference_mode() (Without it, the forward passes are recorded for training and use far more memory.)

### Questions
- Can a small studio use LTX-2.5 for paid client videos? Yes, if the company's yearly revenue is under $10 million; the license allows any purpose under its use rules, and Lightricks claims no rights in your outputs.
- Can a large company test it for free? Yes, for testing, evaluation or non-commercial research outside production; any real use needs the paid license.
- Can I share a version I fine-tuned? Yes, but only under the same license, with a full copy of it included.
- Does it run in ComfyUI? Yes. Lightricks publishes its own ComfyUI nodes and points to them from the main code page.

---
AgentGid's catalog is curated: every entry was checked against the maker's official pages (registry, repository, license and pricing pages) on the date shown; a license value shows the quote it stands on wherever AgentGid has one. Usage numbers come from AgentGid's own scan of public repositories set up for AI coding agents. Claims by makers are reported, not tested by AgentGid.
