AugLy

Why we included this project

Most augmentation libraries pick one data type and stay there; AugLy spans audio, image, text, and video with over a hundred transforms, so a team working across modalities keeps a single API instead of gluing several tools together. The interesting part is its focus on how media actually gets edited online: turning an image into a meme, overlaying emojis, reposting a screenshot. Those are the kinds of changes that break copy detection, hate speech models, and copyright matching, which is exactly what Facebook built it for. It works both ways, expanding training sets and probing where a model fails, and the sub-libraries are split by modality so you only pull in the parts you need.