LTX-2.3 is Great — But the Docs & ComfyUI Examples Need Work
Just want to vent a little — LTX-2/LTX-2.3 is a great model, but the Docs + ComfyUI examples are somewhat... lacking. The Example Workflows should include sample inputs and prompts, and the ComfyUI examples should highlight specific features with use-case demonstrations. Right now, some features you know exist, but it's unclear what they actually do or how to use them. Also, the LoRA examples really should come with visual reference images.
Also, what direction will LTX-2.5 take — will it go the route of simply scaling up the parameter size like 2.3 did, or will it take a new approach like Wan2.2 (just wanted to bring it up) which uses a MoE (Mixture of Experts) architecture split into 2 experts: one handling high-noise stages controlling motion, objects, and anatomy, and another handling low-noise stages for refinement and fine detail — (essentially the high/low noise experts are Wan2.1 models each handled by a separate model) — or will it go in a completely different direction altogether? (Curious what others think about this)