mlsanityData to 7 Oct 2026

Dynamics-Level Watermarking of Flow Matching Models with Random Codes

arXiv 2605.16239 ↗First posted 15 May 2026cs.LGcited by 0

On the map: Diffusion models → Robustness, privacy and adversarial aspects

also in 1 more
Abstract (arXiv)

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.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Robustness, privacy and adversarial aspects, one of 14 directions in Diffusion models. Builds on 8 of its 10 references that are in the map. First posted on arXiv 15 May 2026.

Builds on

8 of 10 references.

Closest ancestors · 2

Same direction, newest first

  1. 2023Tree-Ring Watermarks: Fingerprints for Diffusion Images that are Invisible and Robust31 May
  2. 2023The Stable Signature: Rooting Watermarks in Latent Diffusion Models27 Mar

Other references · 6

  1. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,084
  2. 2022Flow Matching for Generative Modelingcited by 5,426
  3. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flowcited by 3,550
  4. 2018Turning Your Weakness Into a Strength: Watermarking Deep Neural Networks by Backdooringcited by 334
  5. 2017Embedding Watermarks into Deep Neural Networkscited by 227
  6. 2019BlackMarks: Blackbox Multibit Watermarking for Deep Neural Networkscited by 25

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025WaterFlow: Learning Fast & Robust Watermarks using Stable Diffusion
  2. 2024CLUE-MARK: Watermarking Diffusion Models using CLWE
  3. 2025Dynamic watermarks in images generated by diffusion models
  4. 2025Image Watermarking of Generative Diffusion Models
  5. 2026ShapeMark: Robust and Diversity-Preserving Watermarking for Diffusion Models
  6. 2024An Undetectable Watermark for Generative Image Models
  7. 2025A Theoretical Analysis of Discrete Flow Matching Generative Models
  8. 2025MaXsive: High-Capacity and Robust Training-Free Generative Image Watermarking in Diffusion Models