Erase but Preserve: Controllable Removal of Copyrighted Animation Characters via Optimized Semantic Anchors

2026-08-13

AI Summary

This work addresses the challenge of unauthorized generation of copyrighted anime characters in text-to-image diffusion models, where existing methods struggle to precisely and controllably erase highly distinctive and diverse characters without compromising image quality. The authors propose a plug-and-play mechanism operating in the continuous text embedding space: semantic anchors are optimized as character proxies under structural and detail constraints, and a structure-aware adaptive strategy is employed to replace embeddings associated with target characters. This approach enables fine-grained control over erasure intensity, supports multi-target removal, and facilitates model transferability. Extensive experiments demonstrate that the method achieves state-of-the-art performance in both character erasure efficacy and image fidelity, offering superior flexibility, generalization, and overall capability compared to current alternatives.

Abstract

The exceptional generation capabilities of text-to-image diffusion models have raised copyright concerns, particularly the unauthorized reproduction of animation characters. Existing concept erasure methods fall short for animation character erasure: model modification methods struggle to identify suitable anchors for diverse, highly distinctive characters; prompt-based steering methods lack fine-grained control for precise intervention. These approaches often yield incomplete erasure and degraded image fidelity, hindering real-world deployment. In this paper, we propose a controllable method operating on the model's continuous textual representation to erase target characters during generation. We optimizes an anchor embedding via structural and detailed constraints to serve as a character surrogate, then replaces target-related embeddings with the anchor via a structure-aware adaptive strategy. Experiments show that our method achieves state-of-the-art erasure effectiveness and image fidelity preservation, while supporting controllable erasure degree, multi-target removal, and model transferability. Moreover, our optimized anchors are plug-and-play with current model modification baselines to improve their erasure performance.

Problem

• copyrighted animation characters
• concept erasure
• text-to-image diffusion models
• controllable removal
• image fidelity

Innovation

• semantic anchors
• concept erasure
• diffusion models
• controllable removal
• animation characters