LTXV Image To Video

You probably already know LTXV as the local speed king - the open video model that drafts a clip in seconds on consumer VRAM where Wan takes minutes. This node is the other side of that family: the hosted LTX-2 model, called as an API from inside ComfyUI core. No install, no extension, no GPU requirement. The whole point is access to the big hosted model (the 19B LTX-2 with synchronized audio, which is a different beast from the local 13B you can squeeze onto your own card) plus Lightricks' official infrastructure.

The workflow is the same pattern as every partner/video node: you feed it an image as the first frame, write a prompt, and it uploads, generates, and returns a VIDEO output. One input, one output, nothing to download.

Where this node is genuinely opinionated is in its constraints. duration runs 6–20 seconds, but anything over 10 seconds only works on the Fast model at 1920x1080 and 25 FPS - a hard cap that the node enforces with an error if you try to cheat it. resolution tops out at 4k (3840x2160) with 2560x1440 and 1920x1080 below it, and fps is 25 or 50. You also get generate_audio, off by default, which adds AI-generated audio matched to the scene when you turn it on - that's the LTX-2 signature feature, and it's the reason to pick this over a local draft.

If you've used LTX locally, you know the model's quirks carry over: it punishes thin prompts, so spend words on motion and camera behavior rather than "beautiful cinematic." And a single-frame start image means the model invents everything after frame one - it's a draft machine, not a control freak. The natural place for this in a workflow is the same draft-then-refine loop the local community already uses: get the composition and motion rough in LTX, then move the result somewhere more controllable for the final pass. With the API version you're paying per render for that speed, so it suits quick ideation runs more than your 20th iteration of the same shot.

The model dropdown sits above everything, and it's a plain combo listing the hosted models - including the Fast tier that unlocks the long durations. Pick that one for speed, the standard model for quality, and test cheap before you commit to a 4k 50fps render, because the price badge follows the resolution you select.

Two practical notes. Only one input image is supported - don't feed it a batch and expect a montage. And like every hosted node, your frame is uploaded to Lightricks' infrastructure, so keep the sensitive stuff out of this graph. It's been in core since the LTXV API nodes landed in early 2026, and it remains the fastest way to get an LTX-2-quality clip without owning the hardware.