Then when you want to read the image, the viewer on an application level (or maybe server, if it makes sense for cloud storage) takes the unique info, deduces what to fill in, and generates the complete image just-in-time. Maybe someone out there is already working on something like this? 4/4

As someone who takes way too many photos, I bet you could probably go way further and squeeze out even more information and somehow rebuild it later. Imagine how many photos of avocados exist in this universe. I bet you could collapse an image to its core uniqueness... 3/4

Like imagine this is the compression equivalent of dehydrating Trisolarians (from Three Body Problem). The body is 90% water - get rid of the "water" in the image - color might be that water. Compress the image down to a black and white skeleton, then when reading, you regenerate the color. 2/4