Mi-Ripple: Restoring Images Degraded by Iterative AI Editing
Mi-Ripple targets the “digital ripple” artifacts that build up after repeated AI image edits.
The paper describes a restoration workflow that separates periodic grid artifacts from granular texture tied up with image content. It uses spectral notching when the artifact can be isolated, and cleaned-reference regeneration when filtering would damage real detail. The authors report very low residual lightness shifts in fourteen notch-only runs, plus a 45% debris-density reduction in one paired regeneration example. HF Daily Papers' note
The paper describes a restoration workflow that separates periodic grid artifacts from granular texture tied up with image content. It uses spectral notching when the artifact can be isolated, and cleaned-reference regeneration when filtering would damage real detail. The authors report very low residual lightness shifts in fourteen notch-only runs, plus a 45% debris-density reduction in one paired regeneration example. HF Daily Papers' note
score 4