Metraipolis › archive › dc_89e85550e44c
Looped Diffusion Transformer — arXiv:2609.40305
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.40305
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T13:47:55.224Z
- SHA-256 of the archived text
bc0a63e3048a932c70a4c8d97f7b642ca3202c4f7f8b0f03e87125bae7ed9e4f- Stable URL of this page
https://metraipolis.com/documento/dc_89e85550e44c
How to cite
Looped Diffusion Transformer — arXiv:2609.40305 — arXiv (en), https://arxiv.org/abs/2609.40305 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_89e85550e44c (Metraipolis archive dc_89e85550e44c, sha256 bc0a63e3048a932c70a4c8d97f7b642ca3202c4f7f8b0f03e87125bae7ed9e4f)
BibTeX
@misc{dc89e85550e44c,
title = {Looped Diffusion Transformer — arXiv:2609.40305},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.40305},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_89e85550e44c; sha256 bc0a63e3048a932c70a4c8d97f7b642ca3202c4f7f8b0f03e87125bae7ed9e4f},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.40305. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
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 by repeatedly running shared Transformer blocks within each denoising step, effectively increasing computational depth while keeping the parameter count fixed. This looped computation enables iterative refinement of internal representations without explicit reasoning tokens. However, naive looping fails to consistently improve image quality. We trace this problem to weak supervision across intermediate loops and unregulated attention updates that progressively erode local information. To overcome these challenges, we propose Looped Diffusion Transformer (Looped-DiT), which combines deep supervision across intermediate loops with self-modulating attention to stabilize looped feature updates. Under matched-parameter and matched-compute settings, Looped-DiT consistently outperforms non-looped baselines. Notably, a 260M-parameter looped model can surpass a model 6.5x larger across multiple text-to-image benchmarks while requiring 4.9x lower inference compute. Beyond this performance gain, we find that looped computation can offer a more effective form of iterative computation for diffusion models, with increasing loop depth yielding larger gains than adding more denoising steps under a fixed inference budget. Furthermore, deeper loops can progressively correct mistakes made in earlier loops, exhibiting behaviors suggestive of latent reasoning. Together, these results show that looped computation offers a promising way to scale visual generation models. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-01 in cs.LG). Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.40305 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-07T13:18:10.701Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-10-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).