Zirnscape Interactive Relational Map - Nomic Test Case (2023)

Visit http://www.tinyurl.com/zirnscape to check this out!
In the early days of Nomic AI, which currently focuses on text data, I provided 12,000 custom-finetuned images (generated using Embeddings + SD 2.1) to the co-founder, and we released this in early 2023.
It works by calculating CLIP embeddings for each image, then projecting that high dimensional vector space into 2D, in which images that are most similar in "content" are closer together. It is fun to explore.

This is such a cool way to manage large image libraries! I look forward to when they circle back to images in the future. It would be awesome to be able to zoom in and see thumbnails of each image.