Comfy-Org / Comfy-Org/ComfyUI

LTX-2 latent preview support

Open
#11,814 0 comments 4 reactions 0 assignees View on GitHub
Feature
Dominant language
Python
Stars
133k
Forks
15.7k
Avg merge
1d 7h
Merged PRs (30d)
158

Description

### Feature Idea

Add latent preview to LTX-2 to quickly see if the model goes in a totally wrong direction to interrupt early. It's fast but with better samplers like res_2s it might take a few minutes to even finish the first half-res phase. LTX-2 seems to be the only model that can support it but doesn't: https://github.com/Comfy-Org/ComfyUI/blob/5cd1113236b0fb032a51bf9d63ba196a2510b0d4/comfy/latent_formats.py#L412

### Existing Solutions

Most models, especially video gen, support latent preview. I assume it just needs training a randomly initialized tensor using the model's VAE and various sample media (images/videos). But I couldn't find a working script for that.

### Other

_No response_

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.