Restoring color in a black-and-white photo can go sideways fast when the tool nudges details, shifts edges, or invents texture that was never there. The video breaks down how to add believable color while keeping the original photo’s shapes, pixels, and fine features untouched.
Coming to you from Unmesh Dinda of PiXimperfect, this practical video lays out a fast route that still leaves you in control. Dinda starts with a step many people skip: changing the file from Grayscale to RGB so color can exist in the document at all. From there, he shows a method that avoids generative AI and instead uses Photoshop’s Neural Filters as a starting point, specifically the Colorize filter. The point is not to accept the first result as “done,” since it often gets small areas wrong. The value is speed: you get a base in seconds, then spend your time only where the photo actually needs human judgment.
The most useful part of that first approach is what happens after the initial color pass. You build a separate layer set to the Color blend mode and paint corrections on top, sampling nearby tones so changes stay consistent with the photo’s lighting. He also leans on selections so brushwork stays contained, which is where people usually waste time when color spills across edges. There’s a quick tip about sampling from an area rather than a single pixel, which helps avoid noisy, speckled color when you’re trying to fix skin, lips, or fabric. If you have ever watched a “simple” colorization turn into a mess of mismatched patches, this section gives you a cleaner way to steer the result.
Then the video moves into a second method that is more aggressive, but still aims to protect the source photo. The idea is to use Generative Fill with a partner model such as Nano Banana Pro or FLUX, generate a colorized version, and then use the Color blend mode so only color is taken from the AI result. You get the color, but you do not keep the AI’s altered edges, reshaped features, or shifted texture. There’s also a prep step involving the crop tool and a square aspect ratio to reduce misalignment, which matters if you plan to compare the generated layer to the original at high zoom. He demonstrates how to check alignment by flipping visibility and inspecting tiny details like eyes and facial hair, where even small drift is obvious.
Once you see that Color blend mode trick in action, the rest becomes a set of decisions instead of a rigid recipe. If the AI color makes skin too bright or a background too dull, you can correct it with adjustment layers and masks rather than starting over. He shows one targeted fix using a solid color layer in Multiply with a mask, which keeps the correction local instead of muddying the whole frame. He also uses Select Subject to separate person from background, then pushes saturation and contrast on the background only, which keeps the subject from turning cartoonish. The video hints at a fully manual option too, using gradient maps and Hue/Saturation to build color from scratch, which is slower but gives maximum control when you do not want any automated guesswork. Check out the video above for the full rundown from Dinda.
No comments yet