Deep Reward Supervisions for Tuning Text-to-Image Diffusion Models

TL;DR

This paper introduces Deep Reward Tuning (DRTune), a novel method for optimizing text-to-image diffusion models directly with reward signals, improving control and image quality, especially for low-level features.

Contribution

The paper presents DRTune, a new algorithm that supervises the entire sampling process of diffusion models and demonstrates its effectiveness in enhancing image quality and control.

Findings

DRTune outperforms existing algorithms on various reward models.

Deep supervision improves low-level control in diffusion models.

Fine-tuning SDXL 1.0 with DRTune yields images comparable to Midjourney v5.2.

Abstract

Optimizing a text-to-image diffusion model with a given reward function is an important but underexplored research area. In this study, we propose Deep Reward Tuning (DRTune), an algorithm that directly supervises the final output image of a text-to-image diffusion model and back-propagates through the iterative sampling process to the input noise. We find that training earlier steps in the sampling process is crucial for low-level rewards, and deep supervision can be achieved efficiently and effectively by stopping the gradient of the denoising network input. DRTune is extensively evaluated on various reward models. It consistently outperforms other algorithms, particularly for low-level control signals, where all shallow supervision methods fail. Additionally, we fine-tune Stable Diffusion XL 1.0 (SDXL 1.0) model via DRTune to optimize Human Preference Score v2.1, resulting in the Favorable Diffusion XL 1.0 (FDXL 1.0) model. FDXL 1.0 significantly enhances image quality compared to SDXL 1.0 and reaches comparable quality compared with Midjourney v5.2.