| title | Scaling Properties of Text Conditioning in Visual Generation |
|---|---|
| authors | Zilong Chen, Chaorui Deng, Kunchang Li, Hongyi Yuan, Haoqi Fan |
| date | 2026-07-31 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2607.29679 |
| affiliation | ByteDance Seed |
| upvotes | 30 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf code |
| abstract | We study empirical scaling properties for text conditioning in visual generation. Such properties have rarely been measured because diffusion loss does not scale with the number of tokens in natural-language prompts. Surprisingly, we find that the converged diffusion loss scales with the amount of structured language in the prompt. To quantify structured language, we adapt two complementary measures: a white-box likelihood metric (GPG) and a black-box attribute metric (ED). Across controlled training runs, the converged diffusion loss decreases approximately linearly with GPG and follows a power law with ED. Guided by these scaling properties, we improve diffusability by constructing structured prompts with semantic and geometric annotations derived from images, and improve promptability by training a prompter through supervised fine-tuning, cold-start, and verifier-gated on-policy distillation. The resulting system outperforms all evaluated open-weight models on nearly every compositional, reasoning, and world-knowledge benchmark, while matching or surpassing the strongest closed-weight models on most evaluations. |
cite
@misc{chen2026scaling,
title={Scaling Properties of Text Conditioning in Visual Generation},
author={Zilong Chen and Chaorui Deng and Kunchang Li and Hongyi Yuan and Haoqi Fan},
year={2026},
eprint={2607.29679},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2607.29679}
}