How to remove the background from multiple images at once
· 5 min read
Batch background removal means cutting the subject out of many photos in one pass instead of opening each one in an editor. Select a folder of images, run a background-removal tool built on an in-browser segmentation model, and download a PNG per photo with the background replaced by transparency, all without a single file leaving your device.
TL;DR: an in-browser model finds the subject in each photo and makes everything else transparent. The output is always PNG, because JPEG cannot store transparency. One file downloads directly; several download as a zip. Hair, glass, and fine fur edges are the cases most likely to need a manual touch-up afterward.
How does a browser remove a background without a server?
Astral Batch's background remover runs an open-source segmentation model, imgly's background-removal library, directly in the browser tab using WebAssembly. The model was trained to separate a foreground subject from its background and outputs a mask; the tool applies that mask as an alpha channel, so background pixels become transparent instead of being deleted or painted over. Nothing about the file needs to reach a server for that inference to run.
Why is the output always a PNG, even if I uploaded a JPEG?
Transparency is a per-pixel alpha value, and JPEG's compression has no channel to store it. PNG does, so a JPEG or PNG source photo always comes back as a PNG with a transparent background. If you need a solid background color instead of transparency, expect an extra step: place the cutout PNG over a background of your choice in an image editor.
What happens when I process more than one image?
| Files selected | What you get |
|---|---|
| One image | A single PNG download, named <original>_no_bg.png |
| Several images | A zip archive containing one PNG per source file |
| A very large batch on a low-end device | Slower per-image processing; memory is the ceiling, not a file-count limit |
Where does automatic background removal still fail?
- Fine detail like loose hair strands or fur can get a slightly soft or clipped edge.
- Transparent or reflective subjects, such as glass or water, confuse the mask because there is no solid edge to find.
- Low-contrast photos, where the subject and background are close in color and brightness, produce a rougher cutout.
- Busy or cluttered backgrounds behind the subject increase the chance of a stray unmasked patch.
What is the safe way to batch-process a folder of photos?
Keep the originals in a separate folder before running anything, since a cutout cannot be reversed back into its original background. Run one test photo first, especially if the subjects share tricky edges like hair or glass, and only queue the rest once that result looks right. If several results need a touch-up, it is faster to fix the mask in an editor than to re-run the whole batch with different settings, because there are no settings to tune.
Remove backgrounds from images in your browser: no accounts, no uploads, drop in a folder and download a zip of transparent PNGs.
Next: compress images without losing quality: transparent PNGs from a cutout batch are often large; compressing them keeps the site fast.
Frequently asked questions
Is batch background removal free?
Yes. No account, no card, and no per-image limit, since everything runs in your own browser tab rather than on a paid server.
Are my photos uploaded to remove the background?
No. The segmentation model runs locally via WebAssembly, which you can verify by going offline after the page loads and processing a photo anyway.
Why is my result a PNG instead of a JPEG?
Transparency requires an alpha channel, which JPEG does not support. Every background-removal result is a PNG so the transparent area can actually be stored.
Can I process a whole folder of photos at once?
Yes. Select or drop multiple images and each one is processed in turn; more than one result downloads automatically as a single zip file.
Will it perfectly cut out hair and fur?
Usually close, but fine strands are the hardest case for any automatic segmentation model and can come out slightly soft or clipped, so check those results before using them for anything final.