ComfyUI unified resize node keeps image and mask aligned for LTX 2.3
A new ComfyUI custom node merges image and mask resizing into one tool, enforcing divisible-by-32 constraints for LTX 2.3 and SDXL workflows while preventing alignment drift.
A ComfyUI custom node consolidates image and mask resizing into a single tool that addresses alignment drift and resolution constraints in LTX 2.3 and SDXL pipelines. The node, available on GitHub, replaces multiple resize nodes — Image Resize, Image Resize V2, and separate mask handlers — with unified geometry logic that keeps both width and height divisible by 32 when needed, or disables that constraint by setting divisibility to 1.
The tool tackles a recurring workflow problem: masks and images drifting out of alignment after separate resize operations, and one dimension meeting divisibility requirements while the other does not. LTX 2.3 enforces strict latent resolution constraints requiring both dimensions to be multiples of 32. SDXL inpainting workflows face similar alignment issues when masks shift slightly relative to the base image after independent resizing steps. The new node eliminates that entire class of errors by processing both tensors through the same geometry pipeline, supporting five scaling modes (fixed dimensions, multiplier, longer side, shorter side, and total megapixels) with optional aspect ratio preservation and center crop. Tensor-based resizing replaces PIL-based methods, removing mismatch artifacts that appear when different libraries handle the same operation. The repository should appear in ComfyUI-Manager once the merge completes, making installation a one-click process for users running the manager extension.