> ## Documentation Index
> Fetch the complete documentation index at: https://docs.comfy.org/llms.txt
> Use this file to discover all available pages before exploring further.

# LTXVSeparateGeneratedKeyframes - ComfyUI Built-in Node Documentation

> Documentation for LTXVSeparateGeneratedKeyframes node.

## Overview

The LTXV Separate Generated Keyframes node splits the generated keyframes added by LTXV Add Generated Keyframes back out of a sampled latent and removes them from the conditioning. Use it before spatially upscaling the video latent. Do not run LTXV Crop Guides first — it treats generated keyframes as disposable guides and drops them.

## Inputs

| Parameter            | Description                                                                                                                                                                              | Data Type    | Required | Range          |
| -------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------ | -------- | -------------- |
| `positive`           | Positive conditioning that holds the generated-keyframe metadata. The metadata is removed from it on output.                                                                             | CONDITIONING | Yes      | N/A            |
| `negative`           | Negative conditioning that holds the generated-keyframe metadata. The metadata is removed from it on output.                                                                             | CONDITIONING | Yes      | N/A            |
| `latent`             | Video latent that contains the generated keyframes. The keyframes are stripped from it on output.                                                                                        | LATENT       | Yes      | N/A            |
| `keyframes_to_batch` | Return the keyframes as a batch of single-frame latents. Leave off to get them as one multi-frame latent, which is what the latent upsampler and a later Add Generated Keyframes expect. | BOOLEAN      | No       | default: False |

### Notes on Inputs

* `positive` must contain generated-keyframe metadata, otherwise the node raises an error telling you to add them with LTXV Add Generated Keyframes first.
* `latent` must be a plain video latent (a 5D tensor). If the video and audio latents are still combined, split them with Separate AV Latent first.
* The number of tokens per latent frame recorded when the keyframes were added must match the tokens per frame of the supplied `latent`. If the latent was rescaled after the keyframes were added, they no longer line up and the node raises an error — separate them before upscaling the latent.
* The recorded keyframe frame range must fit inside the supplied `latent`, otherwise the node raises an error that the keyframes were recorded against a different latent.
* The recorded guide attention entry index must still exist in the conditioning. If the conditioning was rebuilt after the keyframes were added, the node raises an error.

## Outputs

| Output Name | Description                                                                                                                                                                                                                                                                                              | Data Type    |
| ----------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------ |
| `positive`  | Positive conditioning with generated-keyframe metadata removed.                                                                                                                                                                                                                                          | CONDITIONING |
| `negative`  | Negative conditioning with generated-keyframe metadata removed.                                                                                                                                                                                                                                          | CONDITIONING |
| `latent`    | Video latent with the generated keyframes stripped.                                                                                                                                                                                                                                                      | LATENT       |
| `keyframes` | The peeled keyframes, labeled with generated\_keyframe\_indices and generated\_keyframe\_num\_frames. Feed these to a later Add Generated Keyframes to initialize new slots, or to Generated Keyframes To Guides to pin them as frozen image guides (indices are remapped if the canvas length changed). | LATENT       |

## Notes

* The `keyframes_to_batch` parameter determines whether the keyframes are returned as a batch of single-frame latents or as one multi-frame latent.
* The node ensures that the generated keyframes are removed from the conditioning and the latent before any further processing.
* The `keyframes` output can be used to initialize new slots for generated keyframes or to pin them as frozen image guides.
* The node raises a `ValueError` if the latent does not contain generated keyframes or if the keyframes do not match the expected format.
* The node assumes that the generated keyframes were added using the LTXV Add Generated Keyframes node and that they are compatible with the current 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/LTXVSeparateGeneratedKeyframes/en.md)

***

**Source fingerprint (SHA-256):** `295e49181e87445a1c47b2e9413d95b20585b12e89f26f13129ac5d97f913007`