Representation Conditioning for Generation

Representation-alignment methods effectively accelerate the diffusion transformer training. REPA, for example, uses a pretrained vision encoder to sup…

Continue in Briflio to read, react, comment, and share.