Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

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.

Jonas Henry Grebe, Tobias Braun, Anna Rohrbach, Marcus Rohrbach• 2026

Related benchmarks

TaskDatasetResultRank
Image GenerationMS-COCO
FID5.4
10
Copyrighted character erasureCopyrighted Characters 100 generations SD3
Erasure Success (Stitch)2
9
Nudity ErasureI2P nudity
Unsafe Rate6.77
7
Nudity ErasureT2I-RP pornography
Unsafe Rate19.63
7
Nudity ErasureRAB
Unsafe Rate28.77
7
Nudity ErasureMMA
Unsafe Rate1.7
7
Nudity ErasureP4D
Unsafe Rate16.17
7
Nudity ErasureBasic prompt suite
Unsafe Rate10
7
Image Generation Utility PreservationMS-COCO
FID8.2
7
Concept ErasureT2I-RP (test)
Unsafe Rate50.77
6
Showing 10 of 14 rows

Other info

Follow for update