Rethinking Cross-Layer Information Routing in Diffusion Transformers

2026-05-20

AI Summary

This work addresses the limitations of conventional residual connections in diffusion Transformers, which induce forward information explosion, gradient attenuation, and inter-layer redundancy, thereby constraining generation efficiency and quality. The study introduces Diffusion-Adaptive Routing (DAR)—a novel, learnable mechanism that dynamically aggregates historical sublayer outputs through nonlinear fusion, explicitly treating cross-layer information routing as an independent design dimension. DAR is adaptive to the denoising timestep and operates in a non-incremental manner. It is orthogonal to existing architectural enhancements and seamlessly integrates with modern Transformer improvements such as REPA. On ImageNet at 256×256 resolution, DAR improves the FID of SiT-XL/2 to 7.56 (+2.11) and achieves the baseline’s convergence quality with 8.75× fewer sampling steps, while accelerating early-stage training by up to 2×.

Abstract

Diffusion Transformers (DiTs) have become a de facto backbone of modern visual generation, and nearly every major axis of their design -- tokenization, attention, conditioning, objectives, and latent autoencoders -- has been extensively revisited. The residual stream that governs how information accumulates across layers, however, has been directly inherited from the original Transformer. In this paper, we present a systematic empirical analysis of cross-layer information flow in DiTs, jointly along depth and denoising timestep, and identify three concrete symptoms of traditional residual addition, namely monotonic forward magnitude inflation, sharp backward gradient decay, and pronounced block-wise redundancy. Motivated by this diagnosis, we propose Diffusion-Adaptive Routing (\textsc{DAR}), a drop-in residual replacement that performs \emph{learnable, timestep-adaptive, and non-incremental} aggregation over the history of sublayer outputs. Moreover, the proposed \textsc{DAR} is compatible with many modern Transformer enhancement methods, such as REPA. On ImageNet $256\times256$, \textsc{DAR} improves SiT-XL/2 by $2.11$ FID ($7.56$ vs.\ $9.67$) and matches the baseline's converged quality with $8.75\times$ fewer training iterations. Stacked on top of REPA, it yields a $2\times$ training acceleration in the early stage, suggesting cross-layer information routing as an underexplored design axis in diffusion modeling, one that operates orthogonally to existing representation-alignment objectives. Beyond pretraining, \textsc{DAR} can also be applied during the fine-tuning stage of large-scale T2I models and preserves high-frequency details during Distribution Matching Distillation.

Problem

• Diffusion Transformers
• residual stream
• cross-layer information flow
• information routing
• denoising timestep

Innovation

• Diffusion Transformers
• Cross-layer Routing
• Residual Stream
• Timestep-adaptive Aggregation
• Diffusion-Adaptive Routing