Inlay

4/ Two diffusion architectures were proposed: “One-Tower” with a single Transformer decoder encoding the context and denoising the next concept at once, and “Two-tower” where we separate context encoding from denoising.