Apple MLR presents its latest research on generating high-quality images & videos using a multi-resolution diffusion model. Named the Matryoshka Diffusion Models or MDM, this advanced technique operates directly in pixel space (approx 1024px) w/out the need for VAEs or cascaded models.

Apple MLR presents its latest research on generating high-quality images & videos using a multi-resolution diffusion model. Named the Matryoshka Diffusion Models or MDM, this advanced technique operates directly in pixel space (approx 1024px) w/out the need for VAEs or cascaded models.
https://arxiv.org/abs/2310.15111