⚡ Winnougan LTX Resolution & Frame Calculator

Every LTX-2.3 workflow starts with two rules that make no sense until they bite you: width and height must be divisible by 32, and the frame count must be 8n+1 (9, 17, 25, 33…). That means 1920×1080 is invalid - LTX wants 1920×1088, because 1088 is the nearest multiple of 32 to 1080. Ask for a 4-second clip at 24fps and you get 97 frames, not 96, because 96 isn't on the grid. The Winnougan LTX Resolution & Frame Calculator is the node that does all this arithmetic for you, and it's the first thing you should drop into any LTX-2.3 graph.

This is the flagship of the ComfyUI_WLTX_nodes pack, and it earns it. LTX-2.3 is Lightricks' 22B speed-first video model - by mid-2026 the community's consensus pick for "best video model that can run locally" - but the model renders internally at half your requested resolution before upscaling. That's where this node's draft_width and draft_height outputs come in: they're your requested size halved and snapped, ready for the Stage 1 draft pass that used to need a couple of ComfyMath nodes to compute.

How it works

Pick a preset and the node resolves the resolution, rounds it up to a multiple of 32, computes a half-size draft, and converts your duration_seconds at the chosen fps into a valid 8n+1 frame count - always rounding up, so you never get less footage than you asked for. It also gives you raw_frames (plain seconds × fps, no LTX snapping) for places like a PromptRelay max_frames input that want the literal count.

The inputs that matter

• preset - 11 options, all pre-snapped: LTX 512p/720p/1080p/1440p/4K, portrait variants of the first three, and two squares. 720p is 1280×736, 1080p is 1920×1088. Choose Custom to type your own.
• target_width / target_height - only read when preset is Custom. Anything you enter gets rounded up to ÷32 anyway.
• fps - LTX native is 24; the tooltip suggests 30 for social media.
• duration_seconds - the one number most people actually tweak per run.

That's it. The outputs are the payoff: wire width, height, and frames straight into EmptyLTXVLatentVideo, draft_width/draft_height into the Stage 1 pass, fps_int to your save node, and info into a Show Text node for a full wiring summary - it even flags when it had to snap a value, which is how you discover you typed 1080 and got 1088.

Install

It ships in the WLTX pack, so install the pack once and you get all eight nodes under Winnougan LTX in the menu:

cd ComfyUI/custom_nodes
git clone https://github.com/Winnougan/ComfyUI_WLTX_nodes

Then restart ComfyUI. Or skip the terminal - ComfyUI Manager has it; search ComfyUI_WLTX_nodes. There's no requirements.txt in the repo, so nothing extra to pip install; the pack just needs ComfyUI and your LTX-2.3 weights (the distilled model is the recommended path).

Where people get burned

The raw_frames vs frames distinction trips people up constantly - they're different numbers on purpose, and wiring frames into something that wanted the raw count (or vice versa) produces silent timing drift. Also remember the info output warns you about snapping for a reason: if your prompt says "cinematic 4-second loop" and you got 97 frames, that's working as intended. And a genuinely useful habit - the node's own info text reminds you: keep the same seed across the draft and refinement passes or you'll get two unrelated clips stitched together.

One aside: the name is a mouthful, but you'll stop noticing once it's the only math node in your graph.