Pixel-Space Post-Training of Latent Diffusion Models
This addresses a specific flaw in image generation for applications requiring high visual fidelity, but it is incremental as it builds on existing LDM frameworks.
The paper tackles the problem of latent diffusion models (LDMs) generating imperfect high-frequency details and complex compositions by proposing pixel-space supervision during post-training, resulting in significant improvements in visual quality and flaw metrics while maintaining text alignment.
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.