Abra: Scaling Diffusion Image Training

2026-08-17

AI Summary

研究通过Abra系统探索了图像生成模型的计算最优扩展规律，发现扩散模型需要比语言模型更多的数据来达到最优训练效果。

Abstract

Compute-optimal scaling laws guide the training of frontier language models yet remain largely unexplored for visual generation. We present a systematic scaling law study for text-to-image diffusion models using Abra, a controlled family of flow-matching transformers trained across three orders of magnitude worth of compute ($10^{19}$ to $10^{22}$ FLOPs), reaching significantly larger compute budgets than previous works. We demonstrate that diffusion models scale just as predictably as language models but require far more data to train optimally: compute optimality occurs at approximately $200$ image tokens per parameter, ten times the Chinchilla compute-optimal prescription for LLMs. We show that unlike language models, diffusion models are robust to overtraining and that practitioners should err on the side of more data rather than a larger model. Finally, we show that this predictability extends beyond training loss to generative quality metrics, optimal CFG settings, representation quality, and even the shape of the training curves, which collapse onto a universal form.

Problem

• diffusion models
• scaling laws
• visual generation
• compute-optimal

Innovation

• Scaling Laws
• Diffusion Models
• Compute-Optimality
• Flow-Matching Transformers
• Generative Quality