Remove Background from Image
Remove the background from any photo in seconds. Get a transparent, white or coloured background at full resolution, and fix edges with simple brushes.
- Full resolution
- Transparent PNG
- Credit cost shown first
- Never used to train AI
Remove background
How to remove the background from an image
ToolsMotive’s background remover finds the main subject of your photo, such as a person, a product, a pet, a car or a logo, and removes everything behind it. It runs in your browser, so your photos are never uploaded, and you download the result at the full size of your original photo, free and without a watermark.
- Add your images above: JPG, PNG, WebP, GIF, BMP or AVIF. Add one photo or up to 50 at once.
- Pick the subject type. “Photo” uses AI for people, products and pets. “Logo, signature or plain background” removes a solid background by its colour, which keeps sharp edges on graphics.
- Click Remove background. The first time, your browser downloads the AI model once; after that it’s kept on your device and starts straight away.
- Check and touch up. Drag the before/after slider, then paint with Erase to remove leftovers or Restore to bring back anything the AI missed.
- Choose the finish and download: transparent, white, any colour, a blurred version of the original, or your own background photo, with an optional shadow, crop and size. Save as PNG, WebP or JPG, one by one or as a ZIP.
Everything this background remover does
Most people only need “remove bg, download PNG”. When you need more, it’s already there, and none of it costs credits or needs an account:
| Feature | What it’s for |
|---|---|
| Full-resolution download (remove bg HD) | The cut-out keeps your photo’s pixel size (up to 24 megapixels), so a large product shot stays large |
| Fine-edge refinement | A guided filter re-traces the edge against your full-size photo, so strands of hair and fur follow the real outline instead of a blurry halo |
| Erase and Restore brushes | Fix any spot by hand, with adjustable brush size and undo |
| Before/after slider | Compare the cut-out with the original on the same image |
| Batch of up to 50 images | Same settings for every photo; download one by one or all as a ZIP |
| New backgrounds | Transparent, white, any colour, a blurred copy of the original (portrait look) or your own photo |
| Shadows | A soft drop shadow, or a ground shadow that makes products look placed rather than floating |
| Crop and size presets | Crop to the subject, square 1:1 for marketplaces, 4:5 for Instagram posts or 16:9 for slides and banners, with adjustable spacing |
| Sticker outline | A white border around the subject, ready for stickers and thumbnails |
| Recolour | Turn a logo or signature all white (for dark backgrounds) or all black |
| Logo and signature mode | Removes a plain background by colour, with a tolerance setting and an option to keep white areas inside the subject |
| Edge choices | Fine detail, crisp, soft, tighter (trims halos) or wider (keeps more) |
| PNG, WebP, JPG and mask | PNG and WebP keep transparency; JPG is smallest; the black-and-white mask is for designers and video editors |
| Private by design | Everything runs on your device; nothing is uploaded |
How the AI decides what to keep
The tool uses two open-source AI models and picks the right one for your device automatically:
- Accurate mode: BiRefNet-lite. BiRefNet (“bilateral reference network”) was designed for high-resolution “dichotomous” segmentation, the job of separating an object from its background with fine detail. Its authors describe it in Bilateral Reference for High-Resolution Dichotomous Image Segmentation. We use the lighter version, published under the MIT licence in the ONNX format, running on your graphics card through WebGPU.
- Fast mode: U²-Net-p. On browsers without WebGPU, the tool uses the small version of U²-Net, a salient-object model described in U²-Net: Going Deeper with Nested U-Structure for Salient Object Detection (Apache-2.0 licence). It’s quick on any processor; its edges are less precise, so the tool tells you when it was used.
For every pixel, the model estimates how likely it is to belong to the main subject. Clear subject pixels are kept, clear background is removed, and pixels in between, along hair or soft edges, become partly transparent. The tool then refines that edge at your photo’s full size, so the result is sharper than the model’s own working resolution. Both models run with ONNX Runtime Web, an open-source engine for running AI models in the browser.
Which browsers get accurate mode
Accurate mode needs WebGPU, the browser feature that lets web pages use your graphics card. Current versions of Chrome and Edge on computers support it, and support on phones and in other browsers is growing; Can I use keeps an up-to-date list. If your browser doesn’t have it, the tool switches to fast mode and says so, rather than failing.
What works best, and what needs a check
AI background removal is excellent with clear subjects and less certain where subject and background blend. Knowing which is which saves time:
| Photo | What to expect |
|---|---|
| A person or product against a plain wall | Usually clean in one go |
| Logos and signatures on white paper | Use logo mode for crisp edges; see remove white background |
| Hair, fur and soft edges | Use “Fine detail” edges and accurate mode; check at full size |
| Glass, smoke and see-through objects | Some background may show through; touch up with the brushes |
| Subject the same colour as the background | Edges can drift; paint them back with Restore |
| Several people or objects | All main subjects are usually kept; erase the ones you don’t want |
Options explained
Transparent background (PNG or WebP)
A transparent background has no colour at all: whatever you place the image on shows through. That works because PNG and WebP files store an alpha channel, an extra value for every pixel that says how see-through it is (MDN’s image file type guide lists which formats support it). JPG has no alpha channel, so a JPG always gets a solid background; if you choose JPG with “Transparent”, the tool uses white.
White background
White suits online shops, catalogues and forms. Many marketplaces ask for a plain white background on the main product photo; check your platform’s own image rules for size and margins. Combine white with the square crop and a ground shadow for a clean shop-ready image.
Blurred background
Keeps the original scene but blurs it, so the subject stands out, similar to a portrait-mode photo. Good for profile pictures and social posts where a plain colour would look cut-out.
Your own background photo
Choose “My own image” and add a background photo: the subject is placed on top, and the background is scaled to fill the frame.
Edges
“Fine detail” is best for people and animals. “Crisp” suits products and graphics. “Tighter” trims a thin halo left by a busy background; “Wider” keeps a little more when the edge was cut too close.
Which file to download for where you’ll use it
| Where it’s going | Download |
|---|---|
| A website, a design app or a slide over a coloured background | PNG, transparent |
| A web page where file size matters | WebP, transparent (usually much smaller than PNG) |
| An online shop or a form that wants a plain background | JPG on white, square crop |
| A document you’ll print | PNG, transparent, at full size |
| Video editing or compositing | The mask (white = keep, black = remove) |
| A profile picture | JPG on a colour or blurred background, square crop |
What people remove backgrounds for
Product photos
Consistent shop images: remove the background, put it on white, square crop, ground shadow, all in one batch.
Profile pictures
A clean colour or a soft blur behind you, cropped square.
Signatures
A transparent signature to place on documents. Signature background remover →
Logos
A logo on white turned into a transparent PNG, plus an all-white version for dark backgrounds. Logo background remover →
Presentations
Cut-outs that sit on a coloured slide without a white box; use the 16:9 crop.
Stickers and thumbnails
Add a white sticker outline and a drop shadow for a cut-out look.
Tips for batches of product photos
- Shoot consistently: same light, same distance and a plain background that contrasts with the product make every cut-out cleaner.
- Set the finish once: background, crop, spacing, shadow and format apply to every photo in the batch.
- Check each one: click the thumbnails above the preview to inspect and touch up each photo before you download the ZIP.
- Keep the originals: the cut-outs are new files, so you can redo a photo with a different background later.
- Make files smaller: Compress image shrinks the results further, and Resize image sets exact pixel sizes.
On a phone or computer
The tool works in modern browsers on iPhone, Android, Windows and Mac, with nothing to install. On a phone you can pick photos from your gallery; the brushes work with your finger. The accurate model is about 115 MB and downloads once, so use Wi-Fi the first time. Very large photos are scaled to 24 megapixels so phones don’t run out of memory.
Quality, privacy and fair use
Your photos never leave your device
The AI model runs inside your browser. Your images aren’t uploaded to ToolsMotive or anyone else; only the model itself is downloaded to your device.
Your original image is never changed
The cut-out is a new file. You can go back to the original at any time.
Never used to train AI
Your photos and results are never used to train or improve AI models, ours or anyone else’s.
Use it on images you’re allowed to edit
Removing a background doesn’t change who owns an image. Use your own photos, or images you have permission to edit.
Troubleshooting
| Problem | What to do |
|---|---|
| Part of the subject was removed | Paint it back with the Restore brush |
| Background bits are left around the edges | Choose “Tighter” edges, or use Erase with a small brush |
| A logo lost its white inner parts | Tick “keep white areas inside the subject” in logo mode |
| The notice says fast mode was used | Your browser has no WebGPU; open the page in a current Chrome or Edge on a computer for accurate mode |
| Transparent areas turned white or black | You saved a JPG, or the app you opened it in doesn’t show transparency; download PNG or WebP |
| The first run is slow | The AI model is downloading once; the next photos start immediately |
Remove background: questions and answers
How do I remove the background from a photo?
Add the image above and click Remove background. Check the edges with the before/after slider, pick a background and download. It takes a few seconds per photo.
Is the background remover free?
Yes. It runs on your own device, so there are no credits, no sign-up and no watermark, and downloads are full resolution.
How do I get a transparent background?
Keep “Transparent” as the background and download PNG or WebP. JPG files can’t be transparent.
Does it keep the full image quality (remove bg HD)?
Yes. The result has the same pixel size as your original, up to 24 megapixels, and the edge is refined at that size.
Can I remove backgrounds from many photos at once?
Yes, up to 50 at a time with the same settings, and download them all as a ZIP.
Are my photos uploaded?
No. The AI runs in your browser; your images stay on your device.
Can I fix mistakes?
Yes. Paint with Erase to remove leftovers and Restore to bring parts back, with undo.
Are my photos used to train AI?
No. Your images and results are never used to train or improve AI models.
Sources
- Zheng et al.: Bilateral Reference for High-Resolution Dichotomous Image Segmentation (BiRefNet, arXiv 2401.03407; checked 8 November 2026).
- Qin et al.: U²-Net: Going Deeper with Nested U-Structure for Salient Object Detection (arXiv 2005.09007; checked 8 November 2026).
- Hugging Face: BiRefNet_lite-ONNX model card, MIT licence (checked 8 November 2026).
- ONNX Runtime: ONNX Runtime Web documentation (checked 8 November 2026).
- MDN Web Docs: WebGPU API and Can I use: WebGPU browser support (checked 8 November 2026).
- MDN Web Docs: Image file type and format guide, transparency support by format (checked 8 November 2026).