LTX 2.5 in ComfyUI: eight video workflow options

This tutorial presents an LTX 2.5 workflow in ComfyUI with eight video generation options. The speaker says it can run with as little as 8 GB VRAM using a lightweight distilled GGUF model. LoRAs support image detail and transitions, while a preview node lets users inspect an intermediate result before continuing or changing settings.

The text-to-video example uses a 10-second clip at 24 FPS and 1920 × 1088 resolution. Its sampler processes the initial latent at 50% resolution, then restores it to full resolution through latent upscaling and a detail LoRA. Other examples use a song, a single image, or first and last frames to guide generation.

The lip sync options combine existing audio with either one image or multiple reference frames. With first, middle and last frames, users can switch between custom audio and LTX's native audio generation inside the subgraph. The speaker recommends clips of at least 10 seconds for this option so the model has time to move between keyframes, along with a transition LoRA.

Motion transfer uses a source video to animate a still image, with controls for the source starting point, duration, FPS and resolution. The speaker reports more natural movement and better detail than version 2.3; the tutorial does not establish those improvements as universal results.