Products on a lifestyle backdrop
A studio-lit product on a warm interior photograph is the most common mismatch there is, and the easiest one to fix: it is almost entirely white balance.
A clean edge is not what makes a composite convincing. Your subject carries the colour of the light it was photographed under, and your background carries a different one. Put them together and your eye sees two photographs, not one scene, even when it cannot say why.
It changes the subject and nothing else. A tool with this job shipped here once before and got it wrong: it ran the whole picture through an image-to-image pass, so it redrew the subject instead of adjusting it, and it altered the background it was supposed to be matching. This one adds a measured offset to colour values inside the matte. Every pixel the matte does not cover comes back byte for byte identical, and that is asserted in the test suite rather than promised here.
A straight fade between leaving the subject alone at 0 and applying the whole correction at 100. The correction itself is capped whatever this is set to: the tool will move a colour cast by a believable change of white balance and no further, and it will never drag your subject to the same brightness as the wall behind it, because a person is not as bright as a wall. The numbers it measured and the numbers it applied are both printed with the result.
Choose a cutout and a background to begin.
Result proof
Input
A transparent cutout, and the background you are placing it on
Local pass
Both converted to CIE Lab, four statistics measured on each, the capped difference applied inside the matte and weighted by its alpha
Result
The same cutout, colour-matched to its new background
How to verify it: The result prints what it measured beside what it applied, for all four corrections, so a capped adjustment is visibly a capped adjustment rather than a small one. Judge it on the composited preview at full size, looking at where the subject meets the background rather than at the subject itself: a mismatch reads as an edge between two pictures.
A studio-lit product on a warm interior photograph is the most common mismatch there is, and the easiest one to fix: it is almost entirely white balance.
A subject shot indoors under tungsten reads orange against daylight. The chroma correction is what closes that, and it is capped so the person stays the colour they are.
Match every element to the same background and they agree with each other as a side effect.
Run this first and Contact Shadow second. They fix two independent halves of the same problem, and doing the colour first means the shadow is built under a subject that already belongs.
No. It converts both images to CIE Lab, measures the average lightness, the two chroma axes, the spread of lightness and the direction brightness trends across each frame, and applies the difference to the subject. Nothing was trained, nothing is downloaded, and the same two files at the same setting always produce the same bytes.
The previous one was a mistake and it is worth naming. It captioned your background image, turned that caption into a text prompt, and ran your whole picture back through an image-to-image pass. So it did not adjust your subject, it redrew it, and it changed the background it was supposed to be matching to at the same time. This one only ever adds an offset to colour values inside the matte. Every pixel outside the matte is written back byte for byte, and a test asserts that over a whole frame with no tolerance.
Because that would be a false claim about the world. A person is not as bright as the wall behind them, and a black camera is not as bright as the desk it sits on. Forcing the averages to agree is the standard textbook version of this transform and it produces grey silhouettes. This one applies about a third of any measured brightness difference and caps what survives that, so it corrects an exposure mismatch without flattening your subject into its background.
It fits a plane to the brightness of each image: not a comparison of two halves, which any single dark object would wreck, but a least-squares fit against every pixel. That gives a number for how much brighter the scene gets from left to right and from top to bottom. If the background brightens toward the right and your subject does not, the tool adds a gentle ramp across the subject in the same direction. It is a shading cue and it is deliberately small. It cannot move a highlight from one side of a face to the other, and no arithmetic over a flat image can.
The corrected cutout, on its own, with its transparency and its dimensions unchanged. The preview shows it on your background so you can judge the colour, but the size and position in that preview are arbitrary, because where your subject goes is a composition decision you are making somewhere else. The tool corrects colour; it does not lay anything out.
Usually one of three things, none of them colour. The subject casts no shadow on its new surface, which is what the Contact Shadow tool is for. The two photographs were taken at different focal lengths, so the perspective disagrees and no amount of grading fixes that. Or the subject is genuinely lit from a direction the background has no equivalent of, in which case the honest answer is a different photograph.
No. Both files are decoded by your browser and the statistics are measured and applied in the same tab. There is no upload, no server, and no model fetch. Your background image is read and never changed.
Contact Shadow
The other half of why a composite reads as fake. Also no download.
Background Remover
Where the transparency this tool needs comes from.
Image Editor
Run it on a selection inside one photograph, matching a subject to the frame around it.
Photo Check
Find out whether your photo will cut out cleanly before you start.
No upload, no model download, and your original is never changed.
Match a cutout →