Two colour-based tools that are not the same tool
The Magic Wand and Colour Range both select by colour, and people use them interchangeably. They compute different things.
Magic Wand with Tolerance 32 asks, of each pixel: is this within 32 units of the sampled colour? Yes or no. The answer is binary. Every pixel is either fully selected or not selected, and the boundary between them is a step. The only softening you get is the anti-aliasing checkbox and whatever feather you apply afterwards.
Colour Range with Fuzziness 32 asks: how close is this pixel to the sampled colour? It returns a partial selection: a pixel slightly outside the core range is selected at 60 per cent, one further out at 20 per cent. The result has a gradient built in, which is why it produces far better masks on anything that shades.
That is the practical difference, and it is why Colour Range is the better tool for almost everything except a flat screenshot.
GIMP’s equivalents: Fuzzy Select (the wand) has a Threshold and a Draw mask preview; Select by Colour does the same thing non-contiguously. GIMP’s threshold is also essentially binary, and its soft equivalent is to build the mask from a decomposed channel instead.
The controls in Colour Range that matter
Select > Colour Range in Photoshop and Photopea.
Shift-click to add more samples. One click on a sky selects one blue. A sky runs from pale at the horizon to deep overhead; three or four shift-clicks across it produce a selection that covers all of them. This single habit doubles the tool’s usefulness.
Localized Colour Clusters, with its Range slider, restricts the selection to pixels near where you sampled. Without it, sampling a red dress also selects a red sign on the far side of the frame. With it, you can pick one of two identical red objects.
The preview dropdown. Set it to Greyscale and the dialogue shows the mask you are about to make, in black and white, live. Judge it there, not by the marching ants afterwards.
The preset list includes Reds, Yellows, Highlights, Midtones, Shadows, Out of Gamut and Skin Tones. Skin Tones with Detect Faces on is a surprisingly good starting mask for portrait work, not because it is accurate, but because it is 80 per cent right in two seconds and paintable from there.
Where colour selection is exactly right
- A sky through tree branches. No selection tool traces a thousand twigs. Colour Range on the blue does it instantly, and the partial values along each branch are approximately correct.
- Changing one colour across a whole frame: every red item in a catalogue shot, a brand colour on signage.
- Selecting out-of-gamut pixels before a print conversion, to see what is going to clip.
- Building a starting mask for a luminosity or colour-specific adjustment, where you will paint the rest.
Where it fails, and the mechanism
Colour selection fails when subject and background share colour values, and no setting fixes that, because the tool has no other information. A brown dog on a brown sofa, a grey car against a grey wall, a white shirt against a white wall: the numbers overlap.
The diagnosis is quick: look at the three channels individually. If one of them separates subject from background, build from that channel (the previous block’s technique). If none of them does, colour is not the route and you need geometry: a path, a painted mask, or a model that understands objects rather than values.
The second failure is noise and compression. A JPEG’s 8×8 blocks and a high-ISO file’s speckle both mean that visually uniform areas are numerically scattered, so a colour selection comes out with holes and a ragged edge. Applying a one-pixel blur to the image on a duplicate layer before sampling, or blurring the resulting mask, both help; neither is free of cost.
The free path
GIMP’s Fuzzy Select has a useful feature Photoshop lacks: dragging after clicking adjusts the threshold live, with the mask previewed. It is a faster way to find the right value than typing numbers.
For partial, gradient-aware colour selection, GIMP’s route is Colours > Components > Decompose to HSV, then work on the hue or saturation layer as a mask. More steps, same result, and it teaches you what the tool is doing.
Krita has Select Similar Colour and a Colour Range selection, both adequate.
One caution about recolouring
Colour Range is often the first step in changing an object’s colour. A warning that belongs here rather than later: a colour selection selects pixels that are a colour, including the highlights and shadows on the object, which are the same hue at different lightness. Recolour the selection with a Hue shift and the shading usually survives. Recolour it by painting a flat colour and the object goes dead, because you have replaced its shading with a single tone.
The method that keeps the shading: a Hue/Saturation layer with the selection as its mask, adjusting Hue rather than painting, or a solid colour layer set to Colour blend mode, which takes the hue from the fill and the luminosity from the photograph. That second one is the standard trick and it takes ten seconds.
And the honest limit on all of it: changing a red car to yellow works; changing a black car to yellow does not, because black has no hue to shift and no shading information in the colour channels to preserve. Dark and desaturated objects have to be repainted with the tone rebuilt by hand, which is a job of a different size.
Every lesson here is open. Your progress is recorded once you pass the Design, before Any Software exam.
Check your understanding
An editor uses Colour Range to select a black leather bag so they can recolour it tan, and finds that neither a Hue shift nor a Colour-mode fill produces a convincing tan bag. Why?
Pick the one you would defend. Your answer stays private.
Stuck on something?Ask Neema, your AI tutor