I’m looking for software that can organize hundreds of graduation/event photos into folders for each person.
Some photos contain diplomas with names, which could be used to identify and group the surrounding photos.
Before building my own solution, I’d rather use or adapt something that already exists.
Does anyone know an app, GitHub project, or AI tool that does this or something similar?
Open source, paid, cloud-based — all suggestions welcome!
You can get most of the way with existing tools and only build the last step yourself.
For grouping photos by person, face clustering already works well. Immich (open source, self-hosted) detects faces and groups them into people automatically, and you can name each person. digiKam (open source desktop app) does face detection and recognition too and works on local folders. If you want to script it, InsightFace for face embeddings plus a simple clustering step (DBSCAN or HDBSCAN) gives you one cluster per person. The diploma idea is a nice way to name the clusters. For each cluster, run OCR (Tesseract or PaddleOCR, or a vision model) on the photos and look for the one that shows a diploma, take the name from it, and use that as the folder name. Then copy every photo in the cluster into that folder. Photos with several people can go into each person’s folder.
Two things to watch: group photos and side angles cause most of the mistakes, so plan a quick manual review of clusters before you export. And since this is face recognition on real people, check consent rules for your school or country before you run it.
A practical middle path is to keep the originals untouched and work on copies. First cluster by face, then use the capture time to break ties: photos taken seconds apart at the same spot are often the same moment, even if the angle changes. Treat the diploma name as a hint, not proof, because the name in a photo may belong to a different person than the people around it, so confirm the folder name against a clear face shot before you export. Test the whole flow on 20 to 30 photos first, check how it handles group shots, and only then run the full set. That small test usually saves a lot of manual fixing later.