What stood out
Color is the obvious thing to count. For social avatars, I care more about whether the image still works when it is tiny.
This is a follow-up to my old iOS app icon color experiment. Same basic idea, different surface.
I still grouped the avatars by visible color. Then I added checks for thumbnail punch, circular crop risk, visual style, and avatar twins. The avatar-twin check is useful because different accounts can turn into the same small blob after resizing.
How I fetched the sample
The script uses Bluesky's public actor search and profile APIs for a short list of search terms. It dedupes accounts by DID, downloads the avatar image, and analyzes the circular part of the square.
I picked the search terms manually to get a mix of people, publications, projects, artists, and brands. The current terms are art, developer, writer, journalist, science, game, music, design, photography, climate, book, bluesky.
1800
public avatars processed. The mosaic shows 180 of them so the avatars stay visible.
In the larger sample, the top color buckets were tan 32%, red 22%, blue 16%, gray 10%. The top style buckets were portrait/headshot 47%, soft avatar 27%, color-heavy graphic 15%.