API reference

Background removal API

The service accepts an image, processes it through rembg, and returns a PNG with transparency. Authentication is not required.

Base URL https://shnwazdev-rembgapi.vercel.app

Endpoints

GET /api/health

Returns service health.

GET /model

Returns the default model, allowed models, model descriptions, and supported options.

POST /api/remove

Removes the image background from upload, JSON, form, or raw bytes.

GET /api/remove?url=https://example.com/image.jpg

Downloads a public image URL and returns the processed PNG.

Request formats

Multipart upload

curl -s -F file=@input.jpg \
  https://shnwazdev-rembgapi.vercel.app/api/remove \
  -o output.png

JSON image URL

curl -s \
  -H "Content-Type: application/json" \
  -d '{"url":"https://example.com/input.jpg"}' \
  https://shnwazdev-rembgapi.vercel.app/api/remove \
  -o output.png

JSON base64

curl -s \
  -H "Content-Type: application/json" \
  -d '{"image_base64":"data:image/png;base64,..."}' \
  https://shnwazdev-rembgapi.vercel.app/api/remove \
  -o output.png

Model metadata

curl -s https://shnwazdev-rembgapi.vercel.app/model

The metadata response includes whether an app-side size cap is enabled.

Raw bytes

curl -s \
  -H "Content-Type: image/png" \
  --data-binary @input.png \
  https://shnwazdev-rembgapi.vercel.app/api/remove \
  -o output.png

Options

Name	Values	Default
`model`	`u2netp`, `silueta`	`u2netp`
`alpha_matting`	`true`, `false`	`false`
`only_mask`	`true`, `false`	`false`
`post_process_mask`	`true`, `false`	`false`

Errors

400

Missing image input, invalid URL, blocked private host, invalid base64, or unsupported model.

413

The hosting platform rejected the request before the app could process it.

415

The request content type is unsupported or a remote URL did not return an image.

502/504

The remote image could not be downloaded or timed out.

Limits and notes

There is no app-side image size cap by default.
Remote URLs must be public and return an image/* content type.
Private, loopback, link-local, multicast, reserved, and unspecified hosts are blocked.
Vercel platform request, memory, and runtime limits can still apply.
Cold starts can take longer because Vercel installs heavy ML dependencies at runtime.
Warm function reuse is much faster because the model session and files are cached.