Truetype has a hinting mechnism, but colors may not be in the spec. The design of some of thee emoji sets suggests vector art, but at some point, they were rasterized. The use of an AI to recover a sense of the original strikes me as wasteful. (If only the originals hadn’t been trashed, we wouldn’t need to do this)
Perhaps the AI could be trained to figure out where the bezier control points should be placed.
