I quickly perused StockCake's website to see where their training data/models came from, but couldn't find anything.

Depending on the licenses of the content it was trained on, this could be an elaborate way of "laundering" commercial stock photos and other content.

Of course, the legality of such things is being very actively litigated, will be interesting to see where it lands.