AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Learning to Watermark in the Latent Space of Generative Models

Sylvestre-Alvise Rebuffi, Tuan A. Tran, Valeriu Lacatusu, Pierre Fernandez, Tomáš Souček, Nikola Jovanović, Tom Sander, Hady Elsahar, Alexandre Mourachko

ICML 2026 · Image generation › Autoregressive and latent generation · approach: Robust diffusion watermarking · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
watermark in latent space instead of pixel space (post-hoc, DCAE)post-hoc watermarker · DCAE 512x512Inference time per image on CPU: 63 ms -> 3 ms (20x speedup)Improved
watermark in latent space instead of pixel space (post-hoc, DCAE)post-hoc watermarker · DCAE 512x512Avg bit accuracy (%): 97.78 -> 95.18Hurt
watermark in latent space instead of pixel space (post-hoc, DCAE)post-hoc watermarker · DCAE combined attacksBit accuracy (%): 97.29 -> 84.28Hurt
make embedder input-dependent instead of coverless (ablation)post-hoc latent watermarker · DCAE 512x512Avg bit accuracy (%): 91.81 -> 95.18Improved
apply latent embedder after quantization instead of before (ablation)post-hoc latent watermarker · RAR-XLAvg bit accuracy (%): 90.60 -> 93.96Improved
distill latent watermarker instead of pixel watermarker into latent decoder (no extractor loss) (ablation)DCAE latent decoder · DCAEAvg bit accuracy (%): 51.38 -> 94.41Improved
add extractor loss (weight 0.1) to latent decoder distillation (ablation)DCAE latent decoder from latent teacher · DCAEAvg bit accuracy (%): 94.41 -> 96.77Improved
add strong extractor loss (weight 1.0) when distilling pixel watermarker into decoder (ablation)DCAE latent decoder from pixel teacher · DCAEFID: bit acc 51.38 -> 92.95; FID 10.72 -> 17.08Mixed
add extractor loss (weight 0.1) to latent decoder distillation (ablation)RAR-XL latent decoder from latent teacher · RAR-XLAvg bit accuracy (%): 81.40 -> 90.57Improved
distill pixel watermarker into RAR latent decoder with strong extractor loss (ablation)RAR-XL latent decoder from pixel teacher · RAR-XLAvg bit accuracy (%): 97.27 -> 71.25Hurt
distill latent watermarker into the generative model by fine-tuning on watermarked latentsDCAE diffusion model · DCAEAvg bit accuracy (%) / FID: 95.18 -> 94.78 bit acc; FID 11.42 -> 10.90Improved
distill latent watermarker into the generative model by fine-tuning on watermarked tokensRAR-XL · RAR-XLAvg bit accuracy (%) / FID: 90.60 -> 89.13 bit acc; FID 3.56 -> 3.32Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy