On the Diffusibility of High-Dimensional Latents

Representation Autoencoders (RAEs) enable diffusion models to operate in the feature spaces of pretrained visual encoders. However, many off-the-shelf…

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