Solid Colour Background Remover

Swap the background for a flat colour of your choosing. One image at a time.

Drag & drop an image here

or tap to browse (PNG, JPEG, WebP, SVG)

One image at a time, up to 25 MB

Background

Solid Colour

Fill

#FFFFFF

One image at a time

Load one photograph

Drop a single image with a clear subject. Product shots, headshots and anything with a defined outline are what the model handles best.

Tip: A background that contrasts with the subject gives a cleaner mask than one that shares its colours.

Pick the fill colour

Choose the colour the background becomes. Pure white is the usual choice for marketplace listings; a brand grey or a flat neutral suits directories and slide decks.

Tip: If the image is going onto a coloured page, fill with that page colour rather than white so the edge blends invisibly.

Process and download

The subject is composited over the fill in a single operation and the result comes back as a PNG with no transparency left to worry about.

Tip: Zoom to the edges before you use it. Soft outlines are where any segmentation error shows first.

One-pass compositing

The mask is blended directly over your chosen colour rather than exported as transparency and flattened later, which is what prevents a rim of the original background surviving at the edges.

Any colour you like

The fill is a straight colour value, so you can match a brand palette, a marketplace requirement or the exact background of the page the image is going onto.

Everything stays local

The segmentation model runs in WebAssembly in your browser and the compositing happens on a canvas in the same tab. No upload, no remote processing, no copy retained anywhere.

Single image per run

The model is memory hungry enough that concurrent runs crash phone browsers, so the engine funnels every request through one slot and the page accepts one file at a time.

Consistent results across a set

Running a group of photographs through the same fill colour produces a visually uniform set, which is the usual reason for wanting a flat background in the first place.

Model cached after first use

The weights download once per browser and are reused afterwards, so only the first image of a session pays the loading cost.

Frequently Asked Questions

Why not just export a transparent PNG and flatten it myself?+

You can, but the edge quality is worse. Partially transparent pixels in an exported cut-out still carry a trace of the original background colour, and flattening later blends that trace into your new background. Compositing in one pass avoids the intermediate entirely.

Can I use it for marketplace listings that require pure white?+

Yes, that is the common case. Set the fill to white and the output is a photograph on a genuinely uniform white background rather than the near-white a studio backdrop actually produces.

Why only one image at a time?+

The segmentation network holds several large tensors in memory while it runs, on top of the model weights. Two at once exhausts what a browser tab is permitted on mid-range hardware, so the engine allows exactly one and queues anything else.

Run your set one at a time. The model stays cached between images, so only the first is slow.

Does the output still have transparency?+

No. Every pixel is opaque once the fill is applied, which is the point. If you want transparency instead, use the transparent background remover.

What if the subject picks up a coloured edge?+

That usually means the original background was strongly coloured and the model left some of it in the soft boundary. Cropping tighter before processing helps, and so does choosing a fill closer to the original background tone.

Is my photograph sent anywhere?+

No. Both the segmentation and the compositing run in your browser. The tool continues working with the network disconnected once the model has been cached.

Format Comparison

Feature Specification	solid colour background replacement	Alternative
Number of passes	One, mask straight to fill	Two, cut out then flatten
Edge quality	Blended against the final colour	Risk of a residual halo
Where it runs	Your browser and GPU	A remote server
Images per run	One	Usually batched
Output	Fully opaque PNG	Transparent PNG

Technical Overview & Specifications

Deep dive into solid colour background replacement architecture and processing.

Show Details

Platform Overview

Plenty of places will not accept a transparent image. Marketplace listings want a white background, staff directories want a consistent grey, printed materials want whatever the brand guidelines dictate. Producing that by hand means cutting the subject out and then compositing it onto a filled layer, which is two operations and two chances to leave a rim of the old background behind. This tool does both in one pass: the same ISNet segmentation model that produces a transparent cut-out is used to build the mask, and the mask is composited directly over a flat fill instead of being written to an alpha channel.

Doing it in a single pass is not just convenience. When you export a transparent PNG and then flatten it somewhere else, the edge pixels have already been rounded to whatever the intermediate file could store, and any partially transparent pixel carries a memory of the original background colour with it. Compositing straight from the mask means the soft edge is blended against your chosen colour at full precision, so a subject photographed against green does not end up with a green halo once it lands on white. Like every background tool here, this one processes a single image per run, because the model needs the memory and a phone browser will not survive two at once.

Compositing against a fill

The segmentation step is identical to the transparent tool: ISNet produces a per-pixel confidence map describing how strongly each pixel belongs to the subject. What differs is what happens next. Instead of writing that map into the alpha channel of the output, the compositor fills a canvas with your chosen colour, then draws the masked subject over it. Pixels the model is completely confident about are drawn fully opaque, pixels it is unsure about are blended proportionally, and the fill shows through underneath by exactly the amount the model was uncertain. The soft boundary is therefore computed against the final background rather than against nothing.

This matters most on hair, fabric edges and anything semi-transparent. A pixel that is sixty percent subject and forty percent background contains real colour information from both, and the honest way to render it over a new background is to mix them. Exporting to transparency first and flattening afterwards performs that mix twice, once against the void and once against the destination, which is what produces the faint outline people notice on badly cut-out images. Everything here runs on a canvas in the same worker context that produced the mask, with no intermediate file written and nothing transmitted.