To remove an image background, first decide whether it is a flat colour or a photographic scene. A white background around a logo is a different problem from hair against a busy room. The tool below provides solid-colour removal for simple graphics and an optional beta portrait method for photos of people.
Remove your image background
Remove backgrounds from images online for free. Clear solid-colour backgrounds or try AI portrait cutouts, then download a transparent PNG. Your images stay on your device.
Static JPG, PNG or WebP · up to 20 MB and 8 megapixels.
Click the original preview to pick a colour. Only matching pixels connected to an image edge are removed.
Best for photos of people, not general product or animal cutouts. First use downloads an AI library and model from jsDelivr and Hugging Face; those services receive normal download requests, not your image. Processing stays in this browser. Hair and complex edges may need another editor.
Choose an image to begin.
Download transparent PNGYour finished image stays in this tab. Choose compression settings, then process it below.
Remove a white or solid-colour background
- Choose a static JPG, PNG or WebP image within the displayed limits.
- Select Solid-colour background. White is the initial colour.
- Click a background area in the original preview to sample its colour.
- Start with a low colour tolerance and click Remove background. Increase tolerance slightly if a pale outline remains.
- Check the checkerboard preview and download the transparent PNG.
The tool removes matching regions connected to the image edges. An enclosed white area inside a logo stays intact. If the subject has the same colour as a connected background, increasing tolerance can erase part of it. Work from the original and try a lower value if that happens.
Cut out a person from a photo
Select AI portrait cutout for a photo of a person. The first run downloads a model and browser library from Hugging Face and jsDelivr. Your image pixels are processed locally. This is a portrait model, so results for products, animals and scenery may be poor. Hair, translucent material and overlapping objects can also need manual editing elsewhere.
Check before publishing
Inspect fine edges and make sure important parts of the subject remain. A checkerboard represents transparent pixels; it is not added to the download. View the PNG over both a light and a dark background to spot unwanted outlines. The tool does not include brush-based touch-ups.
Keep transparency when converting
PNG and WebP can retain transparency. JPG cannot: converting a transparent cutout to JPG requires a solid fill. Keep the PNG original of your cutout, then make smaller copies for the destinations that need them.
Replace the background with a colour
Choose Replace with a colour under Output background, pick the new colour and run removal again. The colour fills behind the retained subject using the same removal mask. Check the edges before downloading.
Behind the result
A transparent cutout depends on the alpha channel, not on painting the background white. Use solid-colour mode for edge-connected plain backgrounds and portrait mode for people.
Solid-colour removal: edge-connected flood fill
Starting at the image borders, the tool visits neighbouring pixels and compares their RGB colour with the chosen background colour. Pixels within the tolerance are made transparent by setting alpha to zero. Only connected regions are reached; an enclosed white area inside a logo can remain. Higher tolerance accepts a wider colour difference and can also remove subject pixels connected to an edge.
Portrait mode: a predicted alpha mask
Optional portrait mode uses the MODNet model through Transformers.js and WebAssembly on your device. The model estimates a foreground mask for a person. Its mask is resized to the source dimensions and multiplied into the existing alpha channel. Soft mask values allow partially transparent hair edges. This is a prediction and may fail on difficult hair, glass, shadows or non-portrait subjects.
Why the result is PNG
Alpha is a per-pixel opacity value: zero is transparent and full opacity is solid. PNG can store that channel; JPG cannot. The result is encoded as PNG. Solid-colour mode needs no model download. Portrait mode downloads model and runtime files from external providers on first use, but the image itself is processed locally.