How to remove the background from a photo and get clean edges
A transparent background turns an ordinary photo into something you can drop onto any slide, store page or design. Clean edges depend more on the photo than on the tool.
Updated · 4 min read
Removing a background means deciding, for every pixel, whether it belongs to the subject or to what is behind it. For a logo on a white page that decision is easy. For a person with curly hair in front of a bookshelf it is one of the hardest problems in image processing. Knowing which case you are in saves time and gives better results.
Two ways to do it
Color-based removal starts from the edges of the image and spreads inward while the color stays similar, stopping where it meets a sharp change. It is instant and needs nothing to download. It works very well when the background is uniform or a smooth gradient: product photos on a plain backdrop, logos, screenshots, scanned signatures, documents.
AI-based removal uses a segmentation model trained on many photos to recognise the subject itself — a person, an animal, an object — regardless of what is behind it. It handles complex backgrounds and fine detail such as hair. The model is large, often tens of megabytes, so the first use takes a moment to load.
A good rule: try the fast method first. If the background is busy or the edges look wrong, switch to AI.
Taking a photo that is easy to cut out
- Contrast is everything. A dark object on a light background (or the reverse) gives any method a clear edge to find. A white mug on a white table is hard even for AI.
- Keep the background plain. A sheet of paper or a bare wall behind a product is enough.
- Light the subject evenly and avoid hard shadows: a shadow touching the object is often treated as part of it.
- Leave space around the subject. Objects cut off by the edge of the frame lose their outline.
- Use the sharpest photo you have. Motion blur turns edges into a gradient that no method can separate cleanly.
Getting clean edges
After removing the background, look at the result over different colors — the checkerboard pattern hides problems that a dark or bright background reveals. Common issues and fixes:
- A thin halo of the old background around the subject: retake the photo with more contrast, or use the AI mode, which handles soft edges better.
- Holes inside the subject where its color matched the background: the color method cannot tell them apart; the AI method usually can.
- Hair looks chopped: this is where AI earns its download. Even then, a plain backdrop helps.
Saving the result
Save the cut-out as PNG to keep the transparency. JPG cannot store transparent pixels and will fill them with a solid color. If you want a white background for a marketplace listing, add it deliberately rather than relying on the format to do it.
What it is used for
- Product photos for online stores and marketplaces, which often require a white background.
- Profile pictures and presentation slides.
- Logos and signatures to place on documents.
- Composite images and simple designs.
Where the photo goes
Most background-removal services upload your photo to their servers. That is fine for a product shot; less so for pictures of your children or your ID. When removal runs in your browser, the model is downloaded to your device and the photo is processed there: it never leaves.
Remove a photo background in your browser
Try it nowFrequently asked questions
Why does the AI mode take longer the first time?
Because the model has to be downloaded to your device before it can run. After that your browser caches it, and later uses start much faster.
Can I put a new background behind the subject?
Yes: save the cut-out as PNG and place it over any background in a slide, a design tool or an image editor.
Does it work on screenshots and logos?
Very well, and usually with the fast mode: their backgrounds are uniform, which is the easiest case.
Will the photo quality drop?
The subject keeps its original pixels. Only the background is made transparent, so the result is as sharp as the photo you started with.