> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.ltx.io/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.ltx.io/_mcp/server.

# EU AI Act

> How LTX marks the video and audio it generates, as the EU AI Act requires, and how to detect the mark.

Article 50(2) of the EU AI Act requires AI-generated content to be marked in a machine-readable way. Video and audio generated by the LTX API carry an invisible watermark that identifies them as AI-generated. For our other EU AI Act disclosures, see the [LTX Transparency Hub](https://ltx.io/transparency).

## Invisible watermark

A watermark is a pattern hidden in the content itself. Ours sits in the pixels of every frame and in the audio samples, and is designed to be invisible and inaudible.

A watermark isn't a cryptographic signature. It suggests that a file came from LTX, but it can't prove who made the file or that nobody changed it.

We use two open watermarking models from Meta:

* **Video:** [PixelSeal](https://github.com/facebookresearch/videoseal), from the VideoSeal family.
* **Audio:** [AudioSeal](https://github.com/facebookresearch/audioseal).

### Detecting the watermark

The detector is open source and runs in Python, on your own machine.

1. Install the detector by following the [LTX-2 marking guide](https://github.com/Lightricks/LTX-2/blob/main/packages/ltx-pipelines/docs/marking.md).
2. Download the detector weights from [Hugging Face](https://huggingface.co/Lightricks/LTX-2-Marking) to a local folder.
3. Run the detector on your file, as shown in the marking guide.

> **Warning**
>
> A negative result doesn't prove a file isn't AI-generated: it may have been made with another tool, or the watermark may have been removed.

## FAQ

#### Is detection free?

Yes. The detector is open source.

#### Do I need a GPU?

No. Detection runs on CPU. A GPU is faster.