Calibri: Enhancing Diffusion Transformers via Parameter-Efficient Calibration

TL;DR

Calibri is a novel, lightweight calibration method for Diffusion Transformers that significantly improves generative quality and efficiency by optimizing a small set of parameters using evolutionary algorithms.

Contribution

Introduces Calibri, a parameter-efficient calibration technique for DiTs that enhances performance and reduces inference steps using black-box reward optimization.

Findings

Consistently improves generative quality across models

Reduces inference steps for image generation

Optimizes only ~100 parameters efficiently

Abstract

In this paper, we uncover the hidden potential of Diffusion Transformers (DiTs) to significantly enhance generative tasks. Through an in-depth analysis of the denoising process, we demonstrate that introducing a single learned scaling parameter can significantly improve the performance of DiT blocks. Building on this insight, we propose Calibri, a parameter-efficient approach that optimally calibrates DiT components to elevate generative quality. Calibri frames DiT calibration as a black-box reward optimization problem, which is efficiently solved using an evolutionary algorithm and modifies just ~100 parameters. Experimental results reveal that despite its lightweight design, Calibri consistently improves performance across various text-to-image models. Notably, Calibri also reduces the inference steps required for image generation, all while maintaining high-quality outputs.