Registers Matter for Pixel-Space Diffusion Transformers

Nikita Starodubcev, Ilia Sudakov, Ilya Drobyshevskiy, Artem Babenko, Dmitry Baranchuk

arXiv:2605.16147 · 2026-05-14 공개 · arXiv · PDF

parameter-efficient diffusion-transformers vision-transformers pixel-space generation-quality feature-maps dual-stream-architecture noise-levels

Abstract

Vision Transformers (ViTs) are known to exhibit high-norm patch-token outliers that degrade feature map quality, a problem effectively mitigated by \textit{register tokens}. As diffusion models increasingly adopt transformer architectures and move toward pixel-space training, they become closer in form to ViTs, raising the question of whether register tokens are also useful for Diffusion Transformers (DiTs). In this work, we show that DiTs differ from ViTs in a key respect: they do not exhibit patch-token outliers. Interestingly, register tokens significantly improve convergence and generation quality of pixel-space DiTs. By analyzing intermediate representations, we find that register tokens produce cleaner feature maps at high noise levels, which may contribute to their effectiveness in pixel-space generation. We further observe that recent pixel-space DiT architectures implicitly incorporate register-like mechanisms, which may partially account for their strong empirical performance. Motivated by these insights, we investigate a parameter-efficient dual-stream architecture that specializes processing for register tokens and improves pixel-space generation quality with negligible runtime overhead.

한국어 요약

📋 한 줄 요약

**[Diffusion Transformer / Register]** Pixel-space DiT가 ViT와 달리 token outlier가 없으나 register token이 수렴·생성 품질 크게 향상, 듀얼 스트림 아키텍처 제안.

🎯 핵심 기여도

💡 핵심 아이디어

DiT는 ViT의 token outlier 문제를 보이지 않음에도 register token이 효과적인데, 이는 register가 outlier 완화가 아닌 high noise level에서 깨끗한 feature map 형성에 기여하기 때문이며, 이 통찰이 register token에 특화된 듀얼 스트림 아키텍처 설계로 이어진다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: pixel-space DiT의 내부 메커니즘에 새 통찰 제공, register token의 작동 원리에 대한 이해 갱신, pixel-space 생성 아키텍처 설계 가이드. **한계**: pixel-space DiT 중심 분석으로 latent-space DiT나 다른 확산 아키텍처로의 일반화는 추가 검증 필요.

🚀 실용적 활용