LTX-2 Retake

LTX-2 Retake is the "re-do a chunk" node. Instead of generating a video from scratch, you hand it an existing clip and it regenerates only a chosen slice of it - the shot where the character's face melted, or the audio take that came out wrong. It's Lightricks' segment-regeneration feature, packaged for Runware's cloud so you never touch an LTX checkpoint.

That's the practical win, and it's the part people miss when they see "LTX" and assume this is another text-to-video node. LTX 2 landed in January 2026 and immediately started the "is Wan still the default?" argument in the community, but the open-weights version is a first-frame/last-frame model with time-windowed conditioning. Retake is the editing half of that same family: point at a video, say "start here, last this long, swap the video track" and get back an edited clip.

What you set

video is required - a Runware media UUID or URL string for the clip you're fixing. Then:

• providerSettings.lightricks.startTime and ...duration - the segment to regenerate, in seconds. They're gated toggles: flip each BOOLEAN on and its _value field appears. startTime must sit inside the input video, and duration (2–16s) plus startTime can't exceed the clip's length.
• providerSettings.lightricks.mode - what gets replaced: replace_video, replace_audio, or replace_audio_and_video (default). This is the whole point of the node - you can redo the audio track without touching the picture, or vice versa.
• frameImages - optional image inputs if you want to condition the regenerated segment.
• positivePrompt - description of what the retake should look like/say.
• numberResults - variations to generate; each one is a billed run.

Output is video as a native VIDEO socket, and the usual Runware output plumbing (outputFormat, outputQuality, ttl) applies. The safety.mode dropdown (none/fast/full) controls content checking on the request.

Installing

It's part of the Runware/ComfyUI-Runware pack - install once, and every Runware model node (video, image, audio, text) comes along.

ComfyUI Manager: search Runware in the Custom Nodes Manager, install, restart.

Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

No weights to download; the deps are runware-sdk, pillow, and soundfile. You need a Runware API key from runware.ai/api-keys (ComfyUI Settings → Runware API key, or the RUNWARE_API_KEY env var).

Where people get burned

The biggest gotcha is also the feature: you must have the input video hosted somewhere Runware can read - a URL or a Runware-hosted UUID, not a local file path on your disk. Upload via the platform first, or keep your source clips on a URL you control. And remember the cost model: every retake is a billed cloud generation, so iterate on a short duration segment before you commit to regenerating the whole take.