ReNeg: Learning Negative Embedding with Reward Guidance

TL;DR

ReNeg introduces a reward-guided learning approach for negative embeddings in text-to-image generation, significantly improving quality and human preference alignment by learning optimized embeddings transferable across models.

Contribution

The paper presents ReNeg, a novel end-to-end method that learns negative embeddings guided by a reward model, integrating classifier-free guidance into training for the first time.

Findings

Learned negative embeddings outperform handcrafted ones.

Negative embeddings generalize across models and tasks.

Significant improvements in human preference alignment.

Abstract

In text-to-image (T2I) generation applications, negative embeddings have proven to be a simple yet effective approach for enhancing generation quality. Typically, these negative embeddings are derived from user-defined negative prompts, which, while being functional, are not necessarily optimal. In this paper, we introduce ReNeg, an end-to-end method designed to learn improved Negative embeddings guided by a Reward model. We employ a reward feedback learning framework and integrate classifier-free guidance (CFG) into the training process, which was previously utilized only during inference, thus enabling the effective learning of negative embeddings. We also propose two strategies for learning both global and per-sample negative embeddings. Extensive experiments show that the learned negative embedding significantly outperforms null-text and handcrafted counterparts, achieving substantial improvements in human preference alignment. Additionally, the negative embedding learned within the same text embedding space exhibits strong generalization capabilities. For example, using the same CLIP text encoder, the negative embedding learned on SD1.5 can be seamlessly transferred to text-to-image or even text-to-video models such as ControlNet, ZeroScope, and VideoCrafter2, resulting in consistent performance improvements across the board.