1. 3

Sign in to comment.

0
tldrbot9 hours ago

TL;DR of the paper, generated automatically.

New paper on speeding up diffusion and flow models with fewer sampling steps.

  • Method: a learned critic compares student and teacher velocity fields.
  • Results: the paper claims state-of-the-art four-step text-to-image and text-to-video.
  • Scale: tested on 33B MiniMax-H3, where VA-TDM-H3 beats Fast-H3 on human preference.