osmarks.net research teams performed PCA on the matrix of meme embeddings, generating many axes of varying interpretability. The results can be found here. Some of the more obvious axes were incorporated into the frontend as sliders. Sparse autoencoders are being investigated as an alternative if relevant datasets can be acquired.
It's already far too late.