ClassDiffusion: More Aligned Personalization Tuning with Explicit Class Guidance

TL;DR

ClassDiffusion introduces a semantic preservation loss to fine-tuning diffusion models, effectively maintaining concept integrity and compositional abilities, thereby improving personalized image and video generation.

Contribution

The paper proposes ClassDiffusion, a novel method that explicitly preserves semantic consistency during personalization tuning of diffusion models, enhancing compositional capabilities.

Findings

Semantic preservation loss reduces concept drift.

Improved compositional ability in personalized generation.

Effective extension to personalized video synthesis.

Abstract

Recent text-to-image customization works have proven successful in generating images of given concepts by fine-tuning diffusion models on a few examples. However, tuning-based methods inherently tend to overfit the concepts, resulting in failure to create the concept under multiple conditions (*e.g.*, headphone is missing when generating "a `dog wearing a headphone"). Interestingly, we notice that the base model before fine-tuning exhibits the capability to compose the base concept with other elements (*e.g.*, "a dog wearing a headphone"), implying that the compositional ability only disappears after personalization tuning. We observe a semantic shift in the customized concept after fine-tuning, indicating that the personalized concept is not aligned with the original concept, and further show through theoretical analyses that this semantic shift leads to increased difficulty in sampling the joint conditional probability distribution, resulting in the loss of the compositional ability. Inspired by this finding, we present **ClassDiffusion**, a technique that leverages a **semantic preservation loss** to explicitly regulate the concept space when learning a new concept. Although simple, this approach effectively prevents semantic drift during the fine-tuning process of the target concepts. Extensive qualitative and quantitative experiments demonstrate that the use of semantic preservation loss effectively improves the compositional abilities of fine-tuning models. Lastly, we also extend our ClassDiffusion to personalized video generation, demonstrating its flexibility.