DEV Community

Vipul Singh
Vipul Singh

Posted on

Built a background remover that runs real AI (ISNet) locally — no upload, ever

Nearly every free background-removal tool works the same way: you upload your photo, their server runs a segmentation model, and you get the result back. That means your image — a product shot, a headshot, anything sensitive — passes through someone else's infrastructure.

Built one that skips that entirely:

  • Runs an ISNet segmentation model directly in the browser via ONNX Runtime WebAssembly — the same class of model a server-side API would use, just local
  • Pixel-level precision, including fine edges like hair and fur
  • Before/after comparison slider to check quality before exporting
  • Fast / Balanced / High Quality processing modes
  • Option to replace the background with a solid color instead of transparency
  • Supports PNG, JPG, WEBP, AVIF input up to 50MB; exports as transparent PNG or WEBP
  • Model (~25-35MB) downloads once and caches — works offline after that, no per-image queue

No account, no file limits, no watermark.

🔗 https://www.forgeplug.com/tools/background-remover

Curious if anyone's compared ISNet vs. the models behind paid tools like remove.bg — open to feedback on edge cases where it struggles.

Top comments (0)