AetherCut^™ background remover

AetherCut's background remover runs MediaPipe SelfieSegmentation locally in WebAssembly. The model file (~3 MB) downloads once and stays cached in your browser. Every frame is segmented on your device's CPU and GPU — there is no server-side inference, no upload queue, and no per-second cost.

If you want to confirm this for yourself, open Chrome DevTools (F12) → Network tab → click the Background Remover tool. You'll see the model file download once on first use, and then absolutely nothing else. No frames go out.

This is the same on-device approach a desktop NLE would take, except you don't have to install anything. It works in a tab, on a Chromebook, on a borrowed laptop, on a plane.

How does AetherCut remove a video background?

It runs each video frame through MediaPipe SelfieSegmentation to create a per pixel alpha mask around the person, then composites that person onto a transparent background or a solid color. Color spill suppression can clean edge tinting.

The tool feeds each frame of your clip through MediaPipe's SelfieSegmentation model — the same model Google ships in Meet for blur-background. It produces a per-pixel alpha mask separating the person from everything behind them.

AetherCut then composites the masked person onto either a transparent background (so you can stack the clip on a different scene in your timeline) or a solid color of your choice. You can also chain Color Spill Suppression after this step to clean up the edge tinting that always shows up around hair and shoulders.

Does AetherCut process background removal on my device?

Yes. AetherCut runs the segmentation model on your device using WebGL and CPU instead of uploading footage to a server. That keeps raw clips on your machine, which matters for private or sensitive work.

Most browser-based background removers (Kapwing's, Canva's, Runway's) send your clip to their servers for processing. The model is identical or near-identical; the difference is where the inference runs.

When the inference runs on a server, your raw footage briefly lives on that company's storage. Even with the best privacy policies, the footage exists outside your machine for the duration of the job. For client work, NDA footage, journalism source material, or anything you'd rather not be on someone else's S3 bucket — that's a meaningful difference.

AetherCut runs the same model on your device's WebGL+CPU. Slower than a cloud GPU, faster than not having to wait for an upload. For a 60-second 1080p clip on a 2024 MacBook Air, the full mask job typically completes in 8–15 seconds.

When is AetherCut background removal not the best choice?

It works best for people. Mask quality drops on products, pets, and other nonhuman subjects. Motion blur, fine hair, and low contrast footage can also limit results, where manual rotoscoping or Magic Mask may work better.

MediaPipe SelfieSegmentation is tuned for people. If you need to remove the background from a product shot, a pet, or anything that isn't human-shaped, the mask quality drops noticeably. AetherCut's Magic Mask (Pro, MediaPipe + MobileSAM) handles arbitrary object segmentation better — also fully local.

If your source footage has motion blur on the subject, very fine hair detail, or low contrast against the background, no in-browser model will match a manual rotoscope. For that we'd genuinely recommend After Effects Roto Brush — paid, slow, but the result is what it is.