Remove an image background
Drop one image here, or paste an image with Ctrl/Cmd+V.
One static JPEG / PNG / WebP · 15 MB · 4 MP · 4096px per side
Choose an image to begin.
MODNet is for a single portrait, not general objects. Hair, glass and similar backgrounds may need manual refinement. PNG keeps original transparency.
Images stay on this device. Selecting an image starts processing and loads model assets from this site.
Help & FAQs
Choose once; switch tools freely.
Generate a portrait mask locally, check its edges, and refine before exporting.
How to remove a background
- Choose one static JPEG, PNG or WebP within 15 MB, 4 million pixels and 4096 pixels per side.
- Background removal starts after image validation. You can cancel and use a manual mask, then erase and restore using the image or exact coordinates.
- Review the exported result. Mask edits automatically update the transparent PNG; download when ready.
Mask editing and local limits
Output keeps the oriented original dimensions and original alpha; the file is re-encoded as PNG. The model’s 256-pixel prediction does not recover missing fine detail. No JPEG flattening or remote processing is used.
The brush diameter is 1–256 original-image pixels. Choose Erase or Restore, draw, or enter X/Y integer coordinates. Keyboard arrows move the editing cursor; Shift moves ten pixels and Enter/Space paints. Escape, pointer cancellation or blur rolls back the current stroke. Undo/redo retains at most six masks; resetting restores the initial model or manual mask.
Processing uses 45-second checkpoints and a task-owned WebGL2 runtime. Cancellation, clearing, replacing the input and leaving invalidate late callbacks, release GPU/image resources and remove downloads. GPU capacity varies by browser/device. Unsupported models fail explicitly; manual editing remains available. Files, masks and settings are not saved or sent to analytics.
Common questions
Are images sent to a server?
No. The fixed model and runtime load from this site after image selection and validation; image pixels stay in this page. File references can continue to another same-language image tool, but masks, history and download results remain local to this tool.
Can it perfectly remove hair, glass or shadows?
No. The compact MODNet portrait model predicts a 256×256 portrait mask, resampled to the original image. Fine hair, fur, glass, shadows, similar colors and complex subjects may leave rough edges or unwanted background. Use erase and restore, then inspect the downloaded PNG.
What if my GPU is unsupported?
The tool reports a model/GPU error rather than exporting a guessed result. You may explicitly start a manual mask without downloading the model, erase background areas and export PNG. Restore brings back original pixels and original transparency.