Geometric Erasure by Contrastive Velocity Matching in Rectified Flows
About
While the rapid adoption of multimodal generative models offers immense potential, it has also increased the risks of harmful content synthesis, deepfakes, and copyright infringements. To address these challenges, concept erasure has emerged as a prospective safeguard. However, as the field gradually transitions from U-Net-based diffusion models to Rectified Flow Transformers, erasure research has struggled to keep pace. In this work, we introduce GEM, a simple but highly effective erasure framework for Rectified Flow models. As part of our contribution, we establish a principled bridge between trajectory-based unlearning grounded in Generative Flow Networks and classic teacher-guided erasure: we translate trajectory-based signals into a teacher-guided flow-matching setup that unifies the strengths of both paradigms. Concretely, a teacher provides complementary attraction and repulsion signals that we combine into a single geometric guidance objective, yielding targeted suppression of unwanted concepts while preserving benign generation.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Image Generation | MS-COCO | FID5.4 | 10 | |
| Copyrighted character erasure | Copyrighted Characters 100 generations SD3 | Erasure Success (Stitch)2 | 9 | |
| Nudity Erasure | I2P nudity | Unsafe Rate6.77 | 7 | |
| Nudity Erasure | T2I-RP pornography | Unsafe Rate19.63 | 7 | |
| Nudity Erasure | RAB | Unsafe Rate28.77 | 7 | |
| Nudity Erasure | MMA | Unsafe Rate1.7 | 7 | |
| Nudity Erasure | P4D | Unsafe Rate16.17 | 7 | |
| Nudity Erasure | Basic prompt suite | Unsafe Rate10 | 7 | |
| Image Generation Utility Preservation | MS-COCO | FID8.2 | 7 | |
| Concept Erasure | T2I-RP (test) | Unsafe Rate50.77 | 6 |