Background Remover — transparent PNG cut-outs from any image URL

Cuts the subject out of a photo and hands back a transparent PNG. Built on rembg with six segmentation models, running on CPU — no API key of your own, no GPU, no per-image SaaS subscription.

Paste a list of image URLs. Get back one cut-out file per image, plus a row describing each one.

Input

Field	Type	Default	What it does
imageUrls	array	—	One image per entry. The same URL listed twice is downloaded and processed once. Up to 500 images per run.
model	select	u2net	u2net (general purpose), u2netp (lightweight, fastest), u2net_human_seg (people), isnet-general-use (sharper edges), silueta (small), birefnet-general (most accurate, slowest).
outputFormat	select	png	png (transparent, lossless), webp (transparent, smaller), jpeg (no alpha channel — the subject is flattened onto backgroundColor).
alphaMatting	boolean	false	A second refinement pass over the edge. Worth it for hair, fur and semi-transparent edges; several times the compute, and billed as its own event.
backgroundColor	string	``	Hex colour to place behind the subject instead of transparency, e.g. #ffffff. Empty keeps it transparent.
maskOnly	boolean	false	Returns the black-and-white subject mask instead of the image, for pipelines that do their own compositing.

JPEG cannot store transparency. Asking for one without a backgroundColor flattens the subject onto white rather than failing the run, and the row says so.

Output

The image goes to the run's key-value store — a dataset record is JSON and cannot hold bytes — and each dataset row carries its URL:

index          position of the image in your input list
source_url     the image that was processed
image_url      the cut-out
model          the segmentation model used
output_format  png, webp or jpeg
alpha_matting  whether the refinement pass ran
billed_event   which event this row was charged under
width, height  pixels of the delivered image
output_bytes   size of the delivered file

Billing

Pay per event: one charge per image, plus a small actor-start fee covering the cold start and model load. The unit is the image rather than the pixel because segmentation downscales to the model's own input size — a phone photo and a print-resolution scan cost almost exactly the same, and charging per megapixel would bill for a difference that is not there.

• A URL that does not download is not charged. Nothing of ours ran.
• A repeated URL is charged once.
• alphaMatting is charged as its own event, because it is a different amount of work, not a slower version of the same one.
• A run stops as soon as your spending limit is reached rather than continuing to work unpaid; everything produced before that point is in the dataset.

Models and licences

All six models ship inside the image, so no weights are downloaded during your run. u2net, u2netp, u2net_human_seg and silueta come from U-2-Net (Apache-2.0), isnet-general-use from DIS (Apache-2.0), and birefnet-general from BiRefNet (MIT). rembg itself is MIT.