# LTXV Freeze Latent

The LTXV Freeze Latent node sets the noise mask of a latent to zero, which keeps that latent clean and unchanged while sampling runs. It works on both video and audio latents, so a latent can be frozen before it is concatenated with others or when it should not be denoised at all.

## Inputs

| Parameter | Description | Data Type | Required | Range |
|-----------|-------------|-----------|----------|-------|
| `latent` | Video or audio latent to freeze. Audio is 4D; video is 5D. | LATENT | Yes | N/A |

### Constraints

- The latent must contain a plain tensor. A concatenated audio-video latent is not accepted; it must be split with the Separate AV Latent node first.
- Only 4D latents (audio) and 5D latents (video) are supported. Any other shape causes an error.
- The generated noise mask is created with zeros using the same device as the input tensor. For video latents the mask has shape (batch, 1, frames, 1, 1); for audio latents it has shape (batch, 1, frames, 1).

## Outputs

| Output Name | Description | Data Type |
|-------------|-------------|-----------|
| `latent` | The input latent with an added noise mask of zeros, so it remains clean during sampling. | LATENT |

> This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! [Edit on GitHub](https://github.com/Comfy-Org/embedded-docs/blob/main/comfyui_embedded_docs/docs/LTXVFreezeLatent/en.md)

---
**Source fingerprint (SHA-256):** `d5d228687f0a124644323c0448dcce53ed6eb2224d3f44d1756079b2a71539ca`
