Correlation-Aware and Gaussianity-Preserving Robust Latent Angular Watermarking for Diffusion Models
Theoretically, theoretically, a rigorous characterization of the induced correlation degradation is provided, deriving in closed form the autocorrelation structure of the watermarked latent and proving that correlations are confined to a sparse, structured set of off-diagonal elements with fixed $\pm\pi/4$ values.