Improving Compositional Generation with Diffusion Models Using Lift Scores

TL;DR

This paper presents a new resampling criterion based on lift scores to enhance compositional generation in diffusion models, improving condition alignment without extra training or modules.

Contribution

It introduces an efficient lift score-based resampling method for diffusion models that improves compositional generation without additional training.

Findings

Lift scores significantly improve condition alignment in synthetic and real data.

The method achieves lower computational overhead during inference.

Effective across multiple tasks including 2D, CLEVR, and text-to-image synthesis.

Abstract

We introduce a novel resampling criterion using lift scores, for improving compositional generation in diffusion models. By leveraging the lift scores, we evaluate whether generated samples align with each single condition and then compose the results to determine whether the composed prompt is satisfied. Our key insight is that lift scores can be efficiently approximated using only the original diffusion model, requiring no additional training or external modules. We develop an optimized variant that achieves relatively lower computational overhead during inference while maintaining effectiveness. Through extensive experiments, we demonstrate that lift scores significantly improved the condition alignment for compositional generation across 2D synthetic data, CLEVR position tasks, and text-to-image synthesis. Our code is available at http://rainorangelemon.github.io/complift.