Normalizing Trajectory Models enable exact likelihood in few-step diffusion
October 7, 2026
Normalizing Trajectory Models (NTM) use conditional normalizing flows to model reverse steps in diffusion models. This method enables exact likelihood training while maintaining the efficiency of few-step generation.
HOW THIS AFFECTS YOU
●
researcherYou can use NTM to bridge the gap between efficient few-step sampling and rigorous likelihood frameworks.