Are We Overthinking One-Step Generation?

Key points

• Diffusion and flow matching are now dominant frameworks for generative modeling in continuous domains, but they typically require multiple network evaluations at inference.1
• Motivated by the difficulty of distributional training from scratch, Priming Models are first trained on a reconstruction objective, e.g., pixel-space denoising at a fixed noise level, before undergoing feature-space distributional post-training.1
• Empirically, across class-conditional image generation, text-to-image generation, and world modeling, Priming Models significantly advance the Pareto frontier of training compute vs. generation quality.1
• On ImageNet at $256 \times 256$ resolution, we achieve a state-of-the-art FID of $1.19$ among one-step generators using approximately one-tenth the training compute of prior methods with comparable FID.1

Sources (1)

• [1]Are We Overthinking One-Step Generation?
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 03:14 PM

  In this work, we introduce Priming Models, a simple two-stage framework for training one-step generators.

  Diffusion and flow matching are now dominant frameworks for generative modeling in continuous domains, but they typically require multiple network evaluations at inference.

Extractive summary: sentences quoted from the sources.