Looped-DiT: 260M model beats 6.5x larger rival on text-to-image with 4.9x less inference compute

burny_tech · x · 2026-10-01

A new paper introduces Looped Diffusion Transformer (Looped-DiT), which runs shared Transformer blocks repeatedly at each denoising step instead of stacking more layers.

Authors include Gao Huang (Tsinghua), Ziwei Liu (S-Lab/CUHK), and Dahua Lin.

Related event: Looped-DiT: Small Model Beats 6.5x Larger Ones by Reusing Transformer Blocks(2 posts)→

Original post →

More from Multimodal

Multimodal channel →