Test-Time Scaling of Diffusion Models via Noise Trajectory Search

TL;DR

This paper introduces a novel test-time noise trajectory search method for diffusion models, significantly improving image generation quality by optimizing noise sequences without requiring differentiability.

Contribution

It proposes a relaxation of the MDP framework to enable efficient noise trajectory optimization using an epsilon-greedy search, achieving state-of-the-art results.

Findings

Achieves up to 164% improvement in class-conditioned/image generation scores.

First practical method for non-differentiable reward-based noise trajectory optimization.

Outperforms baselines and matches/exceeds MCTS performance in experiments.

Abstract

The iterative and stochastic nature of diffusion models enables test-time scaling, whereby spending additional compute during denoising generates higher-fidelity samples. Increasing the number of denoising steps is the primary scaling axis, but this yields quickly diminishing returns. Instead optimizing the noise trajectory--the sequence of injected noise vectors--is promising, as the specific noise realizations critically affect sample quality; but this is challenging due to a high-dimensional search space, complex noise-outcome interactions, and costly trajectory evaluations. We address this by first casting diffusion as a Markov Decision Process (MDP) with a terminal reward, showing tree-search methods such as Monte Carlo tree search (MCTS) to be meaningful but impractical. To balance performance and efficiency, we then resort to a relaxation of MDP, where we view denoising as a sequence of independent contextual bandits. This allows us to introduce an $\epsilon$-greedy search algorithm that globally explores at extreme timesteps and locally exploits during the intermediate steps where de-mixing occurs. Experiments on EDM and Stable Diffusion reveal state-of-the-art scores for class-conditioned/text-to-image generation, exceeding baselines by up to $164\%$ and matching/exceeding MCTS performance. To our knowledge, this is the first practical method for test-time noise trajectory optimization of arbitrary (non-differentiable) rewards.