Mi-Ripple: Restoring Images Degraded by Iterative AI Editing

Repeated reference-conditioned image editing tends to accumulate grid-like artifacts and granular noise across generations, a degradation termed 'digital ripple.' Mi-Ripple introduces a diagnosis-guided restoration pipeline that decouples periodic lattice distortions from content-entangled textures. By combining selective spectral notching, structure-preserving smoothing, and cleaned-reference regeneration, the workflow achieves low-distortion filtering (CIELAB lightness residual standard deviation of 0.08–0.44) and cuts downstream debris density by 45%.