Background Removal
Background removal for developers: API, self-hosted, or client-side.
3 min read · RemoveBGAI Team
If you are building background removal into a product rather than using it as a one-off tool, the decision looks different — cost per image, latency, and where the data goes all matter more than they do for a single edit. This is a rundown of the three general approaches and where each fits.
Try the editor
Remove backgrounds from your next image with RemoveBGAI
Remove BackgroundHosted APIs
Services like remove.bg offer a REST API that takes an image and returns a cutout. This is the fastest way to add background removal to an app, since there is no model to manage — you send a request and handle the response. The tradeoff is per-image or per-credit pricing that scales with usage, and every image you process is uploaded to a third party, which matters for products handling user-uploaded photos or anything privacy-sensitive.
Self-hosted, open-source models
Open-source segmentation models (U^2-Net and its variants are common starting points) can be run on your own infrastructure. This removes the per-image API cost and keeps data on servers you control, at the cost of needing to manage GPU or CPU inference yourself, along with the model updates and edge cases a managed API would otherwise absorb.
This is usually the right call for high-volume, cost-sensitive products where the API pricing at scale would exceed the cost of running your own inference.
Client-side, in the browser
Running a background removal model in the browser (via WebAssembly or WebGPU) shifts the compute cost onto the user's device instead of your servers. There is no per-image cost, no upload latency, and no image data ever reaches your backend, which is the strongest privacy posture available. This site's own editor is built this way — the tradeoff is that you inherit the variance of user hardware, so a very old or low-power device will process more slowly than a mid-range one.
Choosing between them
- Prototyping or low volume: a hosted API is the fastest path to something working.
- High volume with cost sensitivity: self-hosting an open-source model usually pays for itself past a certain scale.
- Privacy-sensitive product, or wanting to avoid per-image cost entirely: client-side processing keeps data on-device and has no marginal cost per image.
Try the editor
Remove backgrounds from your next image with RemoveBGAI
Remove BackgroundReader questions
Is there a genuinely free background removal API?
Most hosted APIs offer a limited free tier (a monthly credit allowance) rather than unlimited free use. For unlimited processing without a per-image cost, self-hosting an open-source model or running removal client-side in the browser are the two options that avoid ongoing API fees entirely.
What is a good open-source model to self-host for background removal?
U^2-Net and its lighter variants are common starting points for self-hosted segmentation, along with newer models built specifically for portrait and product matting. The right choice depends on your accuracy needs versus inference speed on your available hardware.
Can background removal really run entirely in a browser?
Yes, using WebAssembly or WebGPU to run a compact model on the user's device. It avoids server costs and upload latency, at the cost of depending on the user's own hardware for processing speed.
Which approach is best for a privacy-sensitive product?
Client-side processing is the strongest option, since the image data never leaves the user's device. Self-hosting is the next best option if client-side is not feasible, since at least the data stays on infrastructure you control rather than a third party.