What is LTX-2 in ComfyUI?
LTX-2 is Lightricks’ audio-video generation model. ComfyUI provides a visual workflow for using it: you load a template, supply the required models and inputs, and generate a clip.
Version note: This guide covers LTX-2.0. The official documentation also lists later LTX versions. Use the documentation and model files for your chosen version rather than mixing files from different releases. This is a documentation-based guide, not a hardware benchmark or hands-on review. ComfyUI’s LTX documentation
Start with an official workflow
- Update your ComfyUI installation.
- Open the template library’s Video section and choose an LTX-2 workflow.
- Follow its model-download prompts. Use the destinations specified by that workflow; don’t guess folders from another tutorial.
- Start with the template’s settings before changing model variants or adding custom nodes.
For a text-to-video workflow, your prompt is the main input. For image-to-video, supply an image as well. ComfyUI publishes examples for both, including distilled variants. Official workflow examples
Choose the checkpoint deliberately
The LTX-2 model card distinguishes the full development model, quantized variants, and a distilled checkpoint. Its distilled configuration uses eight steps and CFG 1. Those values are not a universal recommendation for every LTX workflow.
Match the checkpoint to the template instead of combining a model name from one guide with settings from another. Lightricks’ model card
A simple first prompt
Try one scene with one action before attempting a complicated sequence. For example:
A ceramic cup rests on a wooden table beside a window. A hand slowly places a spoon next to the cup. The camera remains still. Soft daylight falls across the table, with quiet room ambience and a small sound when the spoon touches the wood.
This is an illustrative prompt, not a tested output. Watch the whole clip and listen separately: does the action happen, does the scene remain consistent, and does the sound make sense?
Troubleshooting checklist
- Missing nodes: Check whether your installation is current and whether nodes failed to import during startup. Newly documented core nodes may not yet be in the stable release.
- Missing model: Compare the workflow’s requested filename with the downloaded file and its expected destination.
- GPU-memory errors: Record your GPU, checkpoint, resolution, frame count, and error message before changing settings. This guide does not promise operation on a particular VRAM capacity.
- Unexpected dimensions: The LTX-2 model card specifies dimensions divisible by 32 and frame counts of the form 8n+1; 121 is one such frame count.
Sources: ComfyUI setup notes, LTX-2 model constraints.
Check limitations and licensing before publishing
Generated video may not follow the prompt precisely. Lightricks also warns about bias, inappropriate outputs, and lower-quality audio in some non-speech cases. Review results before using them commercially or presenting them as factual footage.
The weights use an LTX community license; “available to download” is not the same as unrestricted use. Read the license applicable to the exact checkpoint and planned use. Model card and license link
For code-based inference rather than ComfyUI, consult the official Lightricks repository. Its current quick start may target a newer release, so check the version before copying commands.
Before calling this a tested guide
The next useful addition is a real run: exact ComfyUI/model versions, GPU and VRAM, settings, elapsed time, original workflow screenshot, generated clip, and any failures. Until that evidence exists, this article should remain labelled documentation-based.
About the Author
