No Pixel Left Behind: Filling Gaps in Anime Colorization

Small unpainted regions are a persistent nuisance in anime colorization pipelines, especially when artists rely on paint bucket workflows for base fills. Those gaps are easy to miss, but they add up to manual zooming, inspection, and cleanup that slows production.

GapFill is designed around that specific bottleneck. It uses deep learning to suggest fill colors from surrounding flat-color regions, matching the way anime-style images are typically painted. The system is meant to reduce the effort of finding gaps, zooming in, and choosing the right color, rather than replacing the existing workflow outright.

In a user study with 13 professional colorists, GapFill outperformed conventional methods on both performance and usability for gap-filling tasks. The results also point to a useful nuance for tool builders: prediction accuracy alone did not determine whether artists found the system helpful. In ambiguous cases, the right color can depend on context, and adoption may hinge on whether users trust the AI enough to let it assist rather than override their judgment.

For teams building production tools, this is a good reminder that narrow, workflow-specific assistance can be more valuable than broad automation. The exact integration model matters: a tool like this is most likely to stick if it complements existing paint and review habits instead of forcing a new one.