LTXV Node Evolution and Kornia Version Conflicts in ComfyUI

You are exploring ComfyUI’s integration of the LTX-Video model, specifically looking into how its nodes are structured and loaded. You start by examining the workflow architecture to understand the relationship between different latent separation tools.

Understanding the Node Evolution

You ask if LTXV Separate AV Latent is an evolution of the generic Separate AV Latent. The analysis confirms that while they are related, one is a standard node and the other is specifically adapted for the LTX-Video system. This distinction reflects how ComfyUI evolves to support specialized model architectures like LTXV while maintaining backward compatibility.

Resolving Library Conflicts

You then pivot to another critical issue: loading the necessary nodes for LTXV, specifically focusing on the Kornia library. You hit a wall when trying to upgrade to kornia >= 0.8.3, which breaks the import of specific functions like pad. This forces a rollback to kornia==0.8.2 using pip commands in your embedded Python environment.

The Latest Patch and Future Proofing

However, there is an update: the LTXVideo project has patched this by replacing Kornia’s native pad with torch.nn.functional.pad. With a PR from August 2026 already in motion to fix this, you conclude that sticking to Kornia 0.8.2 is the most stable path forward.