Vis-O-Matic image analysis tools

Image Cutter

Isolate any part of an image

1 · Model
No model loaded.
2 · Images
Load a model first. Segmentation needs it before images are of any use.
Drop images or a folder here or use the buttons below · JPG, PNG, WEBP
Working size and memory
The model resizes every image to 1024 px internally, so working above about 2000 px costs memory and time without sharpening the mask. Exports are rebuilt at whatever size you choose here, so a lower working size does not reduce output quality.
3 · Segment
Load a model and at least one image first.
Click what you want to keep. Shift-click or right-click to subtract. The model returns three candidates and shows the strongest.
Press and drag across the image, then release. The model segments what falls inside the rectangle.
The loaded model has no box input, so a rectangle would be ignored. Switch to SAM 2.1 tiny or SAM 3 in step 1 to use box prompts.
The model is asked about a grid of points across the image and every distinct region it returns is collected. More points find smaller things and take longer.
Applies the same prompt to every loaded image. Positions are relative to each frame, so images of different sizes and shapes are treated alike.
Load a model and an image to begin.
4 · Review
Segment something first. Results appear here for checking before anything is written to disk.
5 · Export
Nothing to export yet.
Bounding box trims the transparent margin — fine for ordinary photography. For 360°/wide-angle work choose a whole-frame or fixed-row canvas instead, so image row keeps a fixed relationship to viewing angle across every image.
Diagnostics
A running log of model state, tensor shapes and timings. Useful if something behaves oddly.
ready
Image Cutter · part of Vis-O-Matic, with Image Analyzer and Image Visualizer
Jonathan Cinnamon, Geography, UBC Okanagan· Segmentation by Transformers.js; model weights carry their own licences.· No model loaded
How Image Cutter works

Image Cutter runs a Segment Anything model in your browser to separate a chosen region of an image from the rest. Point at what you want, or let the model propose regions, then review the result and export it.

  • Point — click what you want. Shift-click or right-click to subtract an area. Keep clicking to refine.
  • Box — drag a rectangle around the subject.
  • Everything — the model proposes every region it can find, and you choose which to keep.
  • Batch — apply one prompt to every loaded image at the same relative position.

Models download once from the Hugging Face CDN and are then cached by your browser. Images stay on your computer. The model proposes a mask; you decide whether to accept it, and every export records which model produced it.