PureCC: Pure Learning for Text-to-Image Concept Customization
This work addresses the problem of maintaining the original model's integrity during concept customization for users who want to personalize text-to-image models without degrading their general capabilities.
This paper introduces PureCC, a method for text-to-image concept customization that aims to preserve the original model's behavior and capabilities while learning new concepts. It achieves this by decoupling the learning objective and using a dual-branch training pipeline, resulting in state-of-the-art performance in both model preservation and customization fidelity.
Existing concept customization methods have achieved remarkable outcomes in high-fidelity and multi-concept customization. However, they often neglect the influence on the original model's behavior and capabilities when learning new personalized concepts. To address this issue, we propose PureCC. PureCC introduces a novel decoupled learning objective for concept customization, which combines the implicit guidance of the target concept with the original conditional prediction. This separated form enables PureCC to substantially focus on the original model during training. Moreover, based on this objective, PureCC designs a dual-branch training pipeline that includes a frozen extractor providing purified target concept representations as implicit guidance and a trainable flow model producing the original conditional prediction, jointly achieving pure learning for personalized concepts. Furthermore, PureCC introduces a novel adaptive guidance scale $λ^\star$ to dynamically adjust the guidance strength of the target concept, balancing customization fidelity and model preservation. Extensive experiments show that PureCC achieves state-of-the-art performance in preserving the original behavior and capabilities while enabling high-fidelity concept customization. The code is available at https://github.com/lzc-sg/PureCC.