🌌 ResTool 64x (LTX 2.3)

LTX is the speed tier of local video generation - the model that renders a clip in seconds where Wan takes minutes, which made the "draft in LTX, refine in Wan" workflow a community staple. When you're iterating fast, the last thing you want is a resolution error killing a run that would have taken seconds. ResTool64xLTX is the LTX-specific member of the pack's ResTool family, tuned to LTX 2.3's geometry and training floor.

It shares the 64-divisibility constraint with the Hunyuan tool (hence the name), but its floor is different: 393,216 pixels, the area of 768x512 - the author's chosen safe reference for LTX 2.3. That's a notably cheaper floor than Hunyuan's 720p, which reflects reality: LTX is a lighter model and doesn't need the same canvas to look right. The tool snaps both output dimensions to multiples of 64 and rescues anything below the floor before snapping.

How it works

Identical three-step machine to its siblings. Derive the second dimension from aspect_ratio and base_resolution, check the pixel total against the 393,216-pixel floor, scale up if it's under, then round to the nearest multiple of 64. debug_info gives you the WxH (Div 64) summary; log carries the full trail. The pack pairs this with its LTX-flavored tools elsewhere in the suite - the 8n+1 frame-count rule enforced by VAESafeFramePadder and the LTX Auto Loop Calculator - and this node is the spatial half of that same philosophy: keep everything the DiT expects.

Inputs and outputs

• aspect_ratio - 16:9, 9:16, 1:1, 4:3, 3:4, 21:9, 3:2, 2:3.
• base_resolution (INT, default 1024) - longest side in pixels.

Outputs: width and height INTs (feed your LTX latent node), plus debug_info and log strings.

Installing it

It ships in comfyui-sequential-batcher:

cd ComfyUI/custom_nodes
git clone https://github.com/Meisoftcoltd/comfyui-sequential-batcher

or ComfyUI Manager, search "comfyui-sequential-batcher", restart. Pure arithmetic.

Common issues

LTX's own gotcha is the frame-count side, not the resolution side: the 13B and later models are strict about 8n+1 frame counts, and that's the pack's VAESafeFramePadder job, not this node's. Get the two mixed up - using the padder for resolution or this tool for frames - and nothing works. Also remember the LTX strength is speed; the tool will happily hand you a 2048-pixel canvas and LTX will cheerfully spend seconds on it, but you're paying in quality-per-second versus just rendering at its happy zone and upscaling.