Looped Diffusion Transformer

Yong Xien Chng, Tianyi Chen, Wenwen Tong +7

Improving text-to-image models has traditionally relied on increasing model size or the number of denoising steps. In this work, we explore an alternative way to scale computation…