How this background remover works
This page uses an open, in-browser model (IMG.LY’s background-removal library, which runs ONNX weights in WebAssembly) to estimate a foreground mask and write a PNG with transparency. Your picture is not sent to a CalcWise server. The model itself is loaded from a public CDN into this browser the first time you run the tool; after that, inference stays on your device. That is different from a “magic” slider that only keys out a flat colour, and it is also different from a cloud API that would require a key and would upload the file.
A typical use is a product photo or a headshot taken against a sofa, a kitchen, or an office wall. You want the subject on a white listing background, a slide, or a packshot. After the cut-out finishes, download the PNG and place it on the background you actually need. Keep the original file as well — a mask is an estimate, and you may want to retry with a tighter crop if the first pass leaves a halo.
Results are better when the subject is clearly separated from the backdrop, the edges are in focus, and the file is not tiny. Busy patterns that match a shirt, very similar tones, motion blur, or a group of people can confuse the model. Transparent objects, flyaway hair, and reflections in glasses are the usual hard cases. If the preview looks soft around the edge, try cropping closer to the subject before you run it again, or clean the fringe in any editor that understands alpha channels.
Because the work is on-device, a phone with little free memory may struggle with a 20-megapixel camera file. Resize first with the CalcWise image resizer if the tab becomes slow or the tab crashes. The download is always PNG; this tool does not flatten to JPEG, which cannot store transparency. Nothing here is a claim of studio-retouching quality, and it is not a substitute for a photographer’s clipping path when a catalogue or a print job needs a perfect edge.
If a future version of the model improves hair or hands, this page can point at a newer library version without changing how you use it: choose a file, run the cut-out, check the preview, and download. The honest limit stays the same — a free, local model will miss sometimes, and you should look at the result before you publish it.