Pixel-Space Post-Training of Latent Diffusion Models

TL;DR

This paper introduces a pixel-space supervision method for post-training latent diffusion models, significantly enhancing high-frequency detail preservation and visual quality without sacrificing text alignment accuracy.

Contribution

It proposes a novel pixel-space post-training approach for LDMs, addressing high-frequency detail issues and improving visual quality in image generation.

Findings

Pixel-space supervision improves high-frequency detail preservation.

Enhanced visual quality and flaw metrics in LDMs.

Maintains text alignment quality after post-training.

Abstract

Latent diffusion models (LDMs) have made significant advancements in the field of image generation in recent years. One major advantage of LDMs is their ability to operate in a compressed latent space, allowing for more efficient training and deployment. However, despite these advantages, challenges with LDMs still remain. For example, it has been observed that LDMs often generate high-frequency details and complex compositions imperfectly. We hypothesize that one reason for these flaws is due to the fact that all pre- and post-training of LDMs are done in latent space, which is typically $8 \times 8$ lower spatial-resolution than the output images. To address this issue, we propose adding pixel-space supervision in the post-training process to better preserve high-frequency details. Experimentally, we show that adding a pixel-space objective significantly improves both supervised quality fine-tuning and preference-based post-training by a large margin on a state-of-the-art DiT transformer and U-Net diffusion models in both visual quality and visual flaw metrics, while maintaining the same text alignment quality.