Transparency

How processing works

nobg.eu is a browser-based AI background remover with local image processing. This page explains exactly how image data is handled in practical terms. How Processing Works on nobg.eu is framed for operators who need durable process notes, not a slogan page. Use the sections below as a working brief before you standardize the editor in a team workflow.

Browser inference

Segmentation inference runs in the browser runtime. The core background removal workflow does not require upload-first cloud processing.

Temporary memory usage

Images are processed in session memory and user-controlled browser storage. The editor is not designed as a permanent cloud image library.

WebGPU and WASM fallback

When available, WebGPU acceleration improves runtime speed. If unavailable, a fallback path including CPU/WASM behavior from runtime libraries keeps the workflow functional.

Step-by-step data path

You select a local file. The browser decodes pixels into memory. The segmentation model runs in the web runtime. You inspect and export. Closing the tab ends the working set for that edit. Network is used for application code and model assets; the architectural goal for core cutouts is not “upload photo → remote GPU → download mask.”

Temporary memory versus storage products

Working images live in session-oriented browser memory and storage mechanisms under your control. That is different from vendors that persist project libraries for later billing. Downloads you save become ordinary files on disk—manage them with your own retention rules.

Acceleration without changing privacy shape

WebGPU may accelerate matrices when available. If not, WASM/CPU still completes locally. Faster silicon does not imply a detour through our servers for inference. Always verify claims by watching the network panel during a cutout: expect model/asset traffic, not photo POST bodies for the core edit.

Failure handling as process, not panic

When masks fail, improve lighting or reshoot before chaining three AI passes. Compression artifacts and matched luminance are physical facts. Processing transparency includes telling users when not to trust a pretty preview.

Analytics and ads are separate layers

Consent Mode and optional advertising may load on marketing surfaces according to your choices and policy pages. Those scripts are not the segmentation engine. Read Privacy and Cookie policies for telemetry scope; do not equate dismissing a cookie banner with uploading a photo for cutout.

Team checklist for IT review

Allowlist nobg.eu asset loads if corporate filters strip WASM. Test in a clean browser profile. Record cold/warm timings on representative hardware. Keep a written SOP for export naming. That package usually satisfies security reviewers faster than vague “AI in the browser” slides.

FAQ

Are images permanently uploaded?

The primary editing flow is local image processing. See the Privacy Policy for broader site operations.

Will quality be identical on all devices?

No. Runtime speed and edge quality can vary by browser, hardware acceleration support, and input complexity.

Step-by-step data path — what should I remember?

For How Processing Works, apply the "Step-by-step data path" section above: keep cutout inference local in the browser, zoom-check edges, and export alpha masters before any channel-specific flatten (point 1).

Temporary memory versus storage products — what should I remember?

For How Processing Works, apply the "Temporary memory versus storage products" section above: keep cutout inference local in the browser, zoom-check edges, and export alpha masters before any channel-specific flatten (point 2).

Acceleration without changing privacy shape — what should I remember?

For How Processing Works, apply the "Acceleration without changing privacy shape" section above: keep cutout inference local in the browser, zoom-check edges, and export alpha masters before any channel-specific flatten (point 3).

Failure handling as process, not panic — what should I remember?

For How Processing Works, apply the "Failure handling as process, not panic" section above: keep cutout inference local in the browser, zoom-check edges, and export alpha masters before any channel-specific flatten (point 4).

Analytics and ads are separate layers — what should I remember?

For How Processing Works, apply the "Analytics and ads are separate layers" section above: keep cutout inference local in the browser, zoom-check edges, and export alpha masters before any channel-specific flatten (point 5).