image-masker-by-class — Masking AI Model
Image Masker by Class turns a segmentation or detection result into a clean black-and-white mask from the class names you pick — pixel-precise for segmentation, box-shaped for detectors. Feeds inpainting. CPU, instant.
Details
- CategoryMasking
- Year-1
- License
Inputs & Outputs
- ImageInputimage
Upstream model's output image. For segmentation models (SAM2/SAM3/Maskformer/Segformer/DeepLabV3Plus), this is the colored overlay — its pixels are matched against the per-class colors in the text input to produce a pixel-precise mask. For object detectors (DETR/YOLO), only its width/height are used as the mask canvas size.
- TextInputstring
Line-separated metadata from an upstream segmentation or detection model. Each line carries a class label and (depending on the upstream) optional color tuple, score, and bbox — the parser auto-detects the format.
- ImageOutputimage
Generated mask image by detected class
Parameters
- target_classesstringdefault
Class names (comma-separated)
Comma-separated class names to include in the mask. Must match labels emitted by the upstream model's text output — e.g. 'object_0' from SAM2/3, 'subject' from BiRefNet, 'person' / 'car' from DETR/YOLOS, 'dog' from RF-DETR-Seg (80 COCO classes). Multiple classes are unioned into a single mask. Leave empty to mask nothing (typically a mistake).
- dilationintegerdefault 0range 0–100advanced
Bbox dilation (0-100 px)
Expand each bbox outward by N pixels. Bbox mode only — used with object detectors to avoid edge artifacts when inpainting (e.g. add 10-20px around a removed object). Has no effect when upstream is a segmentation model (pixel-precise output).
- duplication_handlingstringdefault Individualsoptions: Individuals, Merge, Left Most, Top Most, Right Most, +2 moreadvanced
Multi-instance handling
How to combine multiple bboxes of the same class (bbox mode only). 'Individuals' (default) keeps each as a separate mask rectangle. 'Merge' unions them into one rectangle. 'Left/Top/Right/Bottom/Center Most' pick a single instance by position. Use a positional rule when only one instance should be masked.
Tags
Alternatives in Image Masking
- Image Masker by Coord
Build a B&W mask from explicit bbox coordinates.
- Image Masker by Region
Build a B&W mask from line-separated polygons or bounding boxes.
- Manuel Image Masker
Hand-draw a B&W mask with brush tools.
- Mask Grow / Shrink
Grow or shrink a mask by N pixels. Instant.
- Object Picker
Interactive — click segmented objects to build a mask.