None of the captures are scraped data it's all painstakingly manual albeit crowd-sourced work.

You can see per-section captures for a given website with their corresponding colors and gradients, typography information (weights, sizes, line heights, letter spacing, etc), border styles and whole lot more. Under font family information i also capture foundry, designer related information parsed from the file's metadata.

For similarity search it stores visual embeddings for semantic search and glyph similarity for font similarity searches.

There's also a well-defined taxonomy for more coarsely grained searches.