Abstract
Recent representation alignment (REPA) methods accelerate diffusion transformer training by aligning projections of the transformer's hidden states with representations from pretrained visual encoders. In this work, we explore a reverse and complementary direction to REPA: rather than projecting diffusion representations into the encoder's space, we inject encoder representations into the diffusion transformer, allowing them to actively participate in the denoising process. To this end, we introduce \textit{REPresentation Injection} (REPI), a training framework based on a scaffold-to-internalization strategy, in which projected encoder representations initially serve as a temporary scaffold and are then progressively internalized by the diffusion transformer. REPI outperforms REPA across a wide range of backbones and is highly complementary to it: combining the two yields substantial gains over either alone. Notably, with only 160K training steps, REPI + REPA matches vanilla SiT trained for 7M steps, a speedup of over $43.5\times$. Code will be available at https://jeneveuxpas.github.io/REPI
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Fu, H., Chen, J., Zhao, B., Chen, W., Liu, W., Li, Q., & Mao, X. (2026). Scaffold Then Internalize: Representation Injection for Diffusion Transformers. https://omanscience.com/en/articles/scaffold-then-internalize-representation-injection-for-diffusion-transformers
MLA 9
Fu, Han, et al. "Scaffold Then Internalize: Representation Injection for Diffusion Transformers." https://omanscience.com/en/articles/scaffold-then-internalize-representation-injection-for-diffusion-transformers.
Chicago (author–date)
Fu, Han, Jiacheng Chen, Baoquan Zhao, Weidong Chen, Wei Liu, Qing Li, and Xudong Mao. 2026. "Scaffold Then Internalize: Representation Injection for Diffusion Transformers." https://omanscience.com/en/articles/scaffold-then-internalize-representation-injection-for-diffusion-transformers.
Harvard
Fu, H., Chen, J., Zhao, B., Chen, W., Liu, W., Li, Q. and Mao, X. (2026) 'Scaffold Then Internalize: Representation Injection for Diffusion Transformers', Available at: https://omanscience.com/en/articles/scaffold-then-internalize-representation-injection-for-diffusion-transformers.
Vancouver
Fu H, Chen J, Zhao B, Chen W, Liu W, Li Q, et al. Scaffold Then Internalize: Representation Injection for Diffusion Transformers. https://omanscience.com/en/articles/scaffold-then-internalize-representation-injection-for-diffusion-transformers
IEEE
H. Fu, J. Chen, B. Zhao, W. Chen, W. Liu, Q. Li, and X. Mao, "Scaffold Then Internalize: Representation Injection for Diffusion Transformers," https://omanscience.com/en/articles/scaffold-then-internalize-representation-injection-for-diffusion-transformers.