Remove Image Background

Remove image background. Remove background from JPG and PNG images while preserving quality.

What Remove Background does

Remove Background runs an AI segmentation model directly in your browser to separate the main subject of a photo from everything behind it, then deletes the background pixels and replaces them with transparency. The model looks at the whole image and predicts, pixel by pixel, how likely each one is to belong to the foreground subject versus the background — it isn't picking a single flat color to key out, so it works on busy, multi-colored, or gradient backgrounds the way old-school color-key tools couldn't. Because the model and its weights (about 40MB) run locally after the first download, your image is never uploaded anywhere; segmentation happens entirely on your device. After the raw mask comes out, the tool runs a second local pass that erodes the mask edge, rebuilds the true foreground color at the boundary (removing background color bleed), and selectively protects high-saturation, high-contrast detail connected to the subject — this is what keeps thin strands of hair, fur, or fine edges from being clipped off along with the background. The result is exported as a PNG with a genuine alpha channel, so the cut-out subject can be dropped onto any new background, layered in a design tool, or kept transparent for a logo or product shot. What it can't do: invent detail the camera never captured, separate two overlapping subjects of the same color, or perfectly resolve glass, smoke, or motion-blurred edges — those areas are inherently ambiguous to any segmentation model and will show partial transparency or minor color fringing rather than a clean line.

Why use Remove Background

You reach for this when the finished image needs to sit on something other than the background it was shot against — a product photo that has to go on a white e-commerce listing, a headshot that needs a plain color behind it for a badge or ID, or a subject you want to composite into a different scene entirely. It also matters when speed is the constraint: manually tracing a subject with a lasso or pen tool in an image editor takes real time and a steady hand, especially around hair or fuzzy edges, and this tool collapses that into one click while still handling those edges reasonably well thanks to the detail-protection pass. Because processing happens locally in the browser, it's also the right choice when the image is sensitive — a photo you don't want leaving your machine, or you're on a connection where uploading a large file to a server isn't practical. And since the output keeps alpha transparency, it's a natural first step before compositing, before adding a new background color or image, or before dropping a cut-out subject into a collage or template elsewhere on the site.

One-click AI subject/background segmentation, no manual selection
Runs locally — the image never leaves your browser
Edge-detail protection keeps hair, fur, and fine strands intact
Transparent PNG output ready for compositing

How to remove a background

The tool works from a single control on the Remove Background page — there's no menu of settings to configure, since the model decides the mask automatically.

1

Upload the JPG or PNG image you want to process

2

Click the action button (labeled Apply while idle, and Removing background... once it starts)

3

Wait through the two visible phases — model loading (only on first use, 10-30 seconds to fetch the ~40MB model) and Removing background..., shown by the progress bar

4

Download the result as a transparent PNG once processing shows Done

Frequently Asked Questions

No. This tool downloads an AI segmentation model (about 40MB, cached after the first run) and executes it locally in your browser using @imgly/background-removal. The image data is processed on your device and never transmitted anywhere, which is why the first run takes 10 to 30 seconds longer while the model weights download.
The tool accepts JPG and PNG input. Output is always written as PNG, since that's the only common format that preserves the alpha transparency the mask produces — if you need a JPG afterward you'd flatten it onto a background color using another tool first, because JPG has no transparency channel.
Fine strands are the hardest case for any segmentation model because a single pixel can be a genuine mix of subject and background color. The tool runs an extra pass that protects saturated, high-contrast detail connected to the main subject specifically to reduce this, but on very fine or low-contrast hair against a similarly colored background you can still see minor fringing or a slightly muted color where the two blended in the original photo.
Expect trouble with transparent or reflective subjects like glass, water, or smoke, where there's no clean foreground/background boundary for the model to find. Motion-blurred edges, subjects that are the same color as their background, and very low-contrast or low-resolution photos also confuse the segmentation, sometimes cutting off part of the subject or leaving background fragments attached.
This standalone tool always runs the local AI model — fully automatic, no upload, one click. The background removal option inside the full Editor calls a server-side remove.bg API instead, which can give a different (sometimes more precise) cut depending on the image, but requires the image to leave your browser and depends on a configured API key. If you want a manual fallback for a subject the AI struggles with, the Editor also has layer and selection tools for hand-correcting a mask, which this standalone page doesn't offer.
Trim has a 'Remove background first' checkbox that runs the same local model, but its job is different: it uses the mask to find the subject's bounding box and crop to it, discarding transparency and giving you a tightly cropped opaque image. Use Remove Background when you want a transparent cut-out to composite elsewhere; use Trim when you want the photo cropped tight around the subject with the original background gone from the frame's edges.
Not currently. There's no exposed slider for mask tightness, edge feathering, or detail protection strength on this page — Apply runs the segmentation and the edge-refinement pass with a single fixed configuration. If a particular image needs a different balance, your main lever is re-running with a version of the photo that has more contrast or resolution around the subject's edges.
The checkered pattern you may see is just how image viewers and editors visualize transparency; it isn't baked into the file. The PNG contains real transparent pixels via its alpha channel, so opening it in most editors, placing it on a colored background, or using it in a design layout will show through correctly rather than showing checkers.

More Image Tools

Explore other powerful tools for your image editing needs