[awaymazdacx5]

Having token embeddings with diffusion models, for 16x16 transformer encoding. Image is tokenized before transformers compile it. If decomposed virtualization modulates according to a diffusion model.