Training-free Diffusion Model Alignment with Sampling Demons

TL;DR

This paper introduces Demon, a novel inference-time, training-free method for aligning diffusion models with user preferences by controlling noise distribution during denoising, validated through experiments with non-differentiable rewards.

Contribution

Demon is the first inference-time, backpropagation-free approach for diffusion model alignment, enabling preference tuning without retraining or differentiable reward functions.

Findings

Significantly improves aesthetics scores in text-to-image generation.

Works with non-differentiable reward sources like VLM APIs and human judgments.

Easily integrates with existing diffusion models without additional training.

Abstract

Aligning diffusion models with user preferences has been a key challenge. Existing methods for aligning diffusion models either require retraining or are limited to differentiable reward functions. To address these limitations, we propose a stochastic optimization approach, dubbed Demon, to guide the denoising process at inference time without backpropagation through reward functions or model retraining. Our approach works by controlling noise distribution in denoising steps to concentrate density on regions corresponding to high rewards through stochastic optimization. We provide comprehensive theoretical and empirical evidence to support and validate our approach, including experiments that use non-differentiable sources of rewards such as Visual-Language Model (VLM) APIs and human judgements. To the best of our knowledge, the proposed approach is the first inference-time, backpropagation-free preference alignment method for diffusion models. Our method can be easily integrated with existing diffusion models without further training. Our experiments show that the proposed approach significantly improves the average aesthetics scores for text-to-image generation. Implementation is available at https://github.com/aiiu-lab/DemonSampling.