Green/red-list token watermarking survives fine-tuning in autoregressive image models, while latent-diffusion watermarks do not survive the autoencoder or the training noise process.
Generative pre- training from pixels
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Radioactive Watermarks in Diffusion and Autoregressive Image Generative Models
Green/red-list token watermarking survives fine-tuning in autoregressive image models, while latent-diffusion watermarks do not survive the autoencoder or the training noise process.