Vevor Mask Union

ComfyUI nodes for Vevor listing images: union a MASK batch, render translated text into OCR boxes with a real font, square-pad without cropping, and score a rendered locale against the source with a vision model

By rexqin·Created 2 days ago·Updated a day ago· 0

rexqin/comfyui-vevor-mask-union

Nodes—

On cloudLocal install

Stars0

Updateda day ago

Readme

comfyui-vevor-mask-union

VevorMaskBatchUnion — fold a MASK batch into a single union mask.

Why

Apply EasyOCR (ComfyUI-EasyOCR) emits one mask frame per detected text region: a batch of (N, H, W), each frame white only inside its own region.

Core ComfyUI cannot reduce that batch:

MaskComposite raises The expanded size of the tensor (1) must match the existing size (N) when the two masks differ in batch size, so it cannot merge a batch.
ImageCompositeMasked silently slices the mask batch down to the destination's batch size. With a single destination image it keeps frame 0 only and drops the rest — so a 22-region OCR pass composites just 1 region, with no error.

This node fills the gap with torch.amax(mask, dim=0).

Wiring

Apply EasyOCR (MASK) → VevorMaskBatchUnion → GrowMask → FeatherMask → ImageCompositeMasked

Install

cd ComfyUI/custom_nodes
git clone https://github.com/rexqin/comfyui-vevor-mask-union
# restart ComfyUI

Or via ComfyUI-Manager → Install via Git URL.

Notes

(B, H, W) → (1, H, W); a 2D (H, W) input is passed through with a batch dim added.
An empty batch is returned unchanged: there is no size information left to rebuild a frame from.