Inpainting: filling the hole
Removing a person is inpainting — filling a region so it matches its surroundings. The 2000 SIGGRAPH paper “Image Inpainting” propagated lines of equal brightness inward from the edges, as art restorers do [1].
PatchMatch and Content-Aware Fill
Large holes and generative fill
Big removals — a person in the middle of a street — need more context. LaMa (2022) uses a network that sees the whole image at once and was trained on large masks [4]; today's prompt-based editors go further and generate new background, which is why they handle crowds and architecture, and why the fill should be checked for invented details.






































