ComfyUI Unified Image + Mask Resize node forces 32-pixel divisibility for LTX 2.3
A new ComfyUI custom node solves image-mask alignment drift and latent resolution mismatches by forcing both dimensions divisible by 32 in a single resize operation.
A ComfyUI custom node released this week consolidates image and mask resizing into one operation that keeps both dimensions divisible by 32—a requirement for LTX 2.3 and SDXL latent pipelines. The Unified Image + Mask Resize node replaces the usual chain of Image Resize, Image Resize V2, and separate mask-resize nodes that often drift out of alignment mid-workflow.
The node ships five scaling modes: fixed dimensions, multiplier, longer side, shorter side, and total megapixels. All modes support optional aspect-ratio preservation and center-crop logic. The divisibility constraint can be toggled off by setting it to 1, which disables the forced rounding and lets users work with arbitrary dimensions when latent constraints don't apply. The resize logic runs on tensors rather than PIL, which avoids the subtle pixel-shift artifacts that appear when masks and images pass through different backends—a critical detail in inpainting workflows where a one-pixel mask offset can leave visible seams or miss the intended region entirely. The creator built the node to eliminate the "one side divisible by 32, the other not" failure mode that breaks LTX 2.3 and SDXL inpainting when users stack multiple resize nodes without unified geometry. The node is live on GitHub and will appear in ComfyUI-Manager once the pull request merges.