Image Cutout › On-device vs cloud
On-device vs cloud background removal: what actually differs
Last reviewed 2026-09-04.
Short answer. An on-device tool runs the model inside your browser, so the photo never leaves your computer and there is no per-image cost or quota, but you pay a one-time model download and the work is limited by your own hardware. A server-side tool uploads the photo, which makes large batches and heavier models practical, at the cost of sending your image to someone else and usually paying per image or per month. Image Cutout is on-device.
| Dimension | On-device (this tool) | Server-side, typically |
|---|---|---|
| Where the photo goes | Nowhere. It stays in the browser tab. | Uploaded to the provider's servers. |
| Account | None. | Usually required. |
| Cost model | Free, with no quota. | Commonly credits, a per-image fee or a subscription. |
| First use | A one-time model download, then cached. | Nothing to download. |
| Per image after that | About a second to encode on a browser with WebGPU, then milliseconds per click. | An upload, a queue and a download, so it follows file size and connection speed. |
| Works offline | Yes, once the model is cached. | No. |
| Large batches | One image at a time. | Better suited, and often scriptable through an API. |
| Model size ceiling | Bounded by what a browser can download and run. | Bounded by the provider's hardware. |
| Longevity | Keeps working as long as the page loads. | Depends on the service staying available, and on its pricing. |
When on-device is the better choice
- The photo is confidential: client work under NDA, identity documents, medical or legal images, unreleased product shots.
- You want a predictable zero cost for occasional use, with no sign-up.
- You are offline or on an unreliable connection.
- You want to keep full resolution without an upload limit.
When a server-side tool is the better choice
- You need to process hundreds or thousands of images unattended.
- You want it wired into a pipeline through an API.
- You need hair-level alpha matting that a browser-sized model cannot match.
- Your device is old or low-powered and has no WebGPU.
What you give up on-device
Being honest about the trade: a browser-sized Segment Anything model gives clean edges on solid objects and soft edges on hair and fur, and it works at a 1024-pixel internal resolution before the mask is resampled to your photo's real size. A dedicated matting service can do better on fine strands. The measured detail is on the methodology page.
FAQ
Is in-browser background removal as good as a paid service?
For solid subjects such as products, furniture and vehicles the edges are comparable. For fine hair and fur a dedicated alpha-matting service is still better, because a browser-sized model works at a lower internal resolution and has no matting refinement stage.
Does an on-device tool work offline?
Yes, once the model has been downloaded and cached by the browser. The first use needs a connection to fetch the weights.
Why would a free tool have no quota?
Because your own device does the work. There is no per-image server cost to recover, so there is nothing to meter.