[All work](/work?type=work)

# Pinterest Aesthetic Analysis

I used CLIP embeddings to group the images on a Pinterest board, then kept the snapshots so one month's themes could be compared with the next.

[Source code](https://github.com/Slaymish/PinterestBoardEmbedding)

Independent development. Python, CLIP embeddings and clustering.

![Five clusters of six dots around a red centre, spread along an arc.](https://cdn.sanity.io/images/qnuj1c4o/production/2de5feed4b6a1c69eb9e2526478ecfa9f3cc81b4-1600x900.png?w=1280\&q=80\&auto=format)

Clustering gives you labels, and a label isn't an explanation, so the saved assignments keep a route back from a cluster to the images inside it. Collection, embedding, clustering and comparison are separate steps.

The checked-in summary from August 2025 covers 171 images in three clusters. One board, one month.

## State

A working pipeline with saved assignments and a dated comparison.

[All work](/work?type=work)