The question i have,
The image is digital when it is uploaded. There is a fixed assortiment of pixels only differentiated by the contrast of shades of grey.
The instructions say to colorise.
In my head this means every pixel needs to obtain a color value, while still matching contrast with pixels around it.
What ai does instead is provide an entirely newly constructed image made up from entirely new pixels which show “google deepdream” patterns.
This is from the final result where the author was “happier” with the result and asked what i assume is an llm to not invent information that isn’t there. (Zoom in)


This is not what i would call a colorisation, it’s a colorised re-creation. I imagine a real colorisation as simply a layer of color values that combine with the exact pixels of the digital scan. Disabling that layer (using photo software) should return the exact digital original and turning the final colorisation into a black and white filter should still leave it nearly identical to the original. Anything else is a recreation for fun and fantasy purposes only.
To vent, i am so glad we convinced someone in the family not to use the “ai enhanced” version for a loved ones official funeral picture.
Yes this is what A.I. does. It shows you a somewhat glitchy alternate reality where that photo is coloured in instead of black and white. So it might do seemingly strange things like upgrading technology and clothes and stuff to match what people wore when colour photos were a thing.
However, it’s relatively trivial to take the A.I. photo and use it as a colour-only layer in Photoshop, while preserving the original grayscale photograph as the source of truth. It might not line up perfectly but you can smudge it around a bit to fit.


