Commons:WikiVisage
WikiVisage is a tool by Premeditated where users can train an ML model via simple yes/no (or to create boundaries boxes) prompts to recognize specific people, automatically adding P180 (depicts) Structured Data to matching images.
If you have any questions you can ask for help at Commons talk:WikiVisage.
Working on implementing relative position within image (P2677). But have to make sure depicts (P180) works 100% first.
Screenshots
Landing page of WikiVisage in 2026
Example of a project page for WikiVisage in 2026
Limitations
- Can only classify humans (Q5)
- Maximum 9,000 images in total for a project
- Traverses through up to 200 subcategories
- Retrieves a maximum of 1,000 images that have P180 → Person of interest (AKA Bootstrap)
- MAX_CONCURRENT_PROJECTS is set to 3 in total for the worker to process
FAQ
How do I get access?
Everyone has access.
I found an error or I have a suggestion
Add it to Commons talk:WikiVisage.
I lost my entire project! 😭😭😭
Well, it is in beta. Radical changes can happen and could result in wipes. Also, instability is to be expected while in beta.
Why can I only classify human (Q5)?
We have to start somewhere.
Is it mobile friendly?
Yes, kind of. It has PWA support. Best on a computer.
Why is it suuuuper slow??!
It is running on Toolforge with limited resources. Working on tweaking the worker to scale better. Check on Grafana before you complain or k8s-status.
The tool made a lot of bad edits!
You are responsible for what gets sent to Wikimedia Commons. Always review what appears in the "SDC Pending" filters on the project site.
Category:Structured Data on Commons Category:Commons tools