DARASK LTX 2.3 Video Settings

LTX 2.3 asks you to set width, height, frame count and FPS, and unless you've memorized the arithmetic you can't tell what length a clip you're queuing until it's generated. DARASK LTX 2.3 Video Settings fixes that with a live preview badge: the moment you edit any widget, the node recomputes and shows you the real output spec on the node itself - 1024×576 (16:9) · 97f @ 24fps = 4.04s. That's the whole value proposition, and it's genuinely useful because LTX's constraints are unforgiving.

The widgets

Four of them, all required:

• width / height (default 1024×576, step 32) - frame dimensions.
• length (default 97, step 8) - frame count. LTX wants (n × 8) + 1 - 97, 121, 241, etc. The step is set so your increments stay on grid.
• fps (default 24.0) - playback frame rate.

The badge recalculates on every edit: aspect ratio reduced via GCD (1024×576 → 16:9), duration computed from length / fps. No more guessing whether 241 frames at 25fps is 8 or 10 seconds - it's on the node.

Outputs

• width, height, length - wire to EmptyLTXVLatentVideo (and length to LTXVEmptyLatentAudio.frames_number).
• fps - FLOAT, into LTXVConditioning.frame_rate or CreateVideo.fps.
• fps_int - the round()'d integer version, for LTXVEmptyLatentAudio.frame_rate, which wants an INT. This output is the pack's excuse for the DARASK Float → Int node being mostly redundant in new workflows: the rounding is done for you.
• info - the badge text as a string, if you want it logged.

Install and the dependency note

Standard pack install: Manager → search DARASK → install, restart. The node itself is pure math and UI - no models. But it's meant to drive LTX 2.3 workflows, so the nodes it feeds (EmptyLTXVLatentVideo, LTXVConditioning, CreateVideo) come from ComfyUI-LTXVideo, which you'll need installed regardless.

The thing to actually get right

LTX 2.3's constraints are hard - off-grid values fail rather than round (see the LTX ecosystem guide). So use the badge as a check before queueing: if length isn't (n × 8) + 1 or a dimension isn't a multiple of 32, fix it there instead of after a failed generation. The node won't stop you from setting a wrong value - it just shows you the truth fast enough to catch it.