Dynamics-Level Watermarking of Flow Matching Models with Random Codes
This work addresses the need for model provenance and intellectual property protection in generative models, offering a novel watermarking approach that is robust and preserves output distribution.
The paper introduces a dynamics-level watermarking method for flow matching models by embedding a key-dependent perturbation into the velocity field during training, enabling reliable message recovery from black-box queries without degrading generation quality, as demonstrated on MNIST and CIFAR-10.
We introduce a dynamics-level approach to watermarking generative models. Rather than embedding signals into model weights or outputs, we embed the watermark directly into the learned continuous dynamics -- the velocity field of a flow matching model. We formulate this as random coding over a continuous channel: a key-dependent perturbation is added during training, and the message is recovered at detection time from black-box queries. The perturbation is designed to leave the generated distribution unchanged. Experiments on MNIST and CIFAR-10 across different architectures confirm reliable message recovery, preserved generation quality, and chance-level decoding accuracy without the secret key.