Image Transition

Inverse colour lab

A is the original. B has an unknown colour treatment. Estimate a mapping from B back to A, then take it apart.

01 / The image pair

Use the same scene and framing. If aspect ratios differ, B is centre-cropped to match A before fitting.

A · OriginalChoose imageB · TransformedChoose image

JPG, PNG, WebP, AVIF · up to 30 MB / 24 MP each. Files stay in this browser.

02 / Recover the mapping

Fitting scope: whole image. The mapping is applied to all of B.

Model

Selective fitting learns different corrections in different RGB colour ranges, with smooth blending between them. Auto tries this when a simpler fit leaves significant error.

Tolerate small pixel misalignments

Uses 3 × 3 local colour averages for fitting only. Your corrected image is not blurred or spatially warped.

03 / Keep the correction

PNG keeps B’s alpha and uses its matching crop at source resolution. The 33³ LUT is an estimated sRGB mapping; colours absent from this pair are extrapolated.

What can—and can’t—be recovered?

The correction depends on the input colour: different colour ranges can receive different changes. Selective mode fits a colour-local RGB grid. It does not identify the original order of edits, and it never copies A into the corrected output.

The same input colour still gets the same correction at every position. Spatial masks that treat identical colours differently in different places need separate area fits.

Clipped highlights, crushed shadows, greyscale conversion, or multiple original colours collapsed into one cannot be uniquely inverted. Local masks, changed content, and large misalignments may leave visible error.

All comparisons use browser-decoded sRGB pixels. Embedded profiles, compression, and resampling can affect the match. This is not a lossless HDR or colour-profile recovery tool.

Focus the fit on original A

Whole image

Choose an area on A to fit only its colours and the matching pixels in B. This changes the fitting samples—not the output crop or the inspection zoom.

All opaque areas of the image are available for fitting.

A · OriginalTarget colours

Add the original image

B · TransformedUnknown treatment

Add the transformed image

Recovered BEstimated inverse applied

Run the inverse fit to inspect this view

Remaining errorBlack = match · brighter = larger RGB error

Run the inverse fit to inspect this view

A → ? → B

The treatment is unknown. Add the pair and fit B → A to reveal the estimated correction, per-channel error, and spatial breakdown.

Image decomposition

Focus the fit on original A