LTX ComfyUI Workflows

If you are weighing LTX comfyui workflows, the thing that sets LTX apart is that it is a genuinely open video model that also generates sound. LTX is Lightricks' diffusion-transformer family, and the current LTX-2 line was released with open weights, so you can run it on your own hardware instead of renting it by the second. The workflows on this page are the ready-made graphs for that: load an LTX model, give it a prompt or an image, and generate a clip with audio already aligned to the picture.

LTX covers the usual modes and a few extras. Text-to-video turns a prompt into motion, image-to-video animates a still, and keyframe conditioning lets you set frames the clip should pass through, including a last-frame interpolation in the newer release. LTX-2 produces synchronized audio and video together at up to 4K and up to 50 frames per second, with clips up to around twenty seconds. The LTX-2.3 point release sharpens detail with a redesigned VAE, cleans up the audio, and adds native portrait video at 1080 by 1920 for vertical formats.

VRAM is the main thing to plan around. The full 22B LTX-2.3 model documents a 32GB requirement, but FP8 and community GGUF builds bring that down so capable consumer cards can run it, and a distilled variant finishes in about eight steps for faster iteration. Because the graph is open, you control the steps, the resolution, and the conditioning, and you can re-run until the motion reads right.

Everything here is open and editable. Run an LTX workflow on Comfy Cloud with nothing to install when you do not have the VRAM, or download it and run it locally. LTX-2 ships under the LTX-2 Community License, which permits research and commercial use by companies under 10 million dollars in revenue at no cost, and there is no watermark on the output.