Beyond Text Prompts: Precise Concept Erasure through Text-Image Collaboration
This work addresses the need for precise and faithful concept erasure in text-to-image models, enabling safer and more controllable generation for practitioners concerned with unsafe or undesirable content.
TICoE introduces a text-image collaborative erasing framework that precisely removes target concepts from text-to-image models while preserving unrelated content, outperforming prior methods in both concept removal precision and content fidelity.
Text-to-image generative models have achieved impressive fidelity and diversity, but can inadvertently produce unsafe or undesirable content due to implicit biases embedded in large-scale training datasets. Existing concept erasure methods, whether text-only or image-assisted, face trade-offs: textual approaches often fail to fully suppress concepts, while naive image-guided methods risk over-erasing unrelated content. We propose TICoE, a text-image Collaborative Erasing framework that achieves precise and faithful concept removal through a continuous convex concept manifold and hierarchical visual representation learning. TICoE precisely removes target concepts while preserving unrelated semantic and visual content. To objectively assess the quality of erasure, we further introduce a fidelity-oriented evaluation strategy that measures post-erasure usability. Experiments on multiple benchmarks show that TICoE surpasses prior methods in concept removal precision and content fidelity, enabling safer, more controllable text-to-image generation. Our code is available at https://github.com/OpenAscent-L/TICoE.git