--- license: other license_name: ltx-2-community-license-agreement license_link: https://github.com/Lightricks/LTX-2/blob/main/LICENSE tags: - comfyui --- # 22th February 2026: Model loading change Starting from this commit https://github.com/Comfy-Org/ComfyUI/commit/f266b8d352607799afb4adf339cdfa854025185e Embedding connector has been moved from text encoder to diffusion model, so to continue using single files you now have to update KJNodes and use my loaders: ![image](https://cdn-uploads.huggingface.co/production/uploads/63297908f0b2fc94904a65b8/rZmv6UeooRKVVH75BQduJ.png) for GGUF: ![image](https://cdn-uploads.huggingface.co/production/uploads/63297908f0b2fc94904a65b8/3HUaruomWUqOEuNOpP5WR.png) # 22th January 2026: Tiny VAE and live previews: LTX2 Tiny VAE has been trained by [madebyollin](https://github.com/madebyollin), and now has preliminary support in ComfyUI nightly version. It can be used with normal VAE Loader and encode/decode nodes, but the quality is very low so it's only useful for preview purposes. Also currently for live animated sampler preview it can be only used with my **LTX2SamplingPreviewOverride** -node in KJNodes, simply load the VAE and plug it in, this overrides any preview setting too. # 13th of January 2026 update !!IMPORTANT!! Turns out the video VAE in the initial distilled checkpoints has been wrong one all this time, which (of course) was the one I initially extracted. It has now been replaced with the correct one, which should provide much higher detail ## requires using updated KJNodes or new enough core ComfyUI VAELoader to work correctly --- Separated LTX2 checkpoint for alternative way to load the models in Comfy ![image](https://cdn-uploads.huggingface.co/production/uploads/63297908f0b2fc94904a65b8/7rNUxIEgfdX1rx0adyJCV.png) LTX2 GGUFs that include the metadata also work with the setup, if your GGUF nodes are up to date. Dev and distill embeddings_connector weights are extracted from their respective original models, they do give quite different results, but it's unclear if there's a "correct" one to use in any given situation. --- Loras folder includes rank reduced loras that can be useful to reduce the memory requirements.