Miyang's Mi-Ripple Targets 'Digital Ripple' in Iterative AI Edits

A three-author paper from Miyang Technology in Shanghai gives a name to something anyone chaining reference-conditioned AI edits will have noticed: 'grid-like and granular textures, commonly described as digital ripple.' The Hugging Face preprint, by Jiayin Chen, Yicheng Xu and Muting Wang, calls its cleanup system Mi-Ripple and pitches it as a 'diagnosis-guided restoration workflow.'

Mi-Ripple 'separates periodic lattice artifacts from content-entangled granular texture, then combines selective spectral notching, structure-aware smoothing, and cleaned-reference regeneration,' the abstract says. That split lets the pipeline notch out artifacts when they sit in an isolated part of the spectrum, and regenerate the image when notching would erase legitimate detail.

The reported numbers are small. Across fourteen notch-only executions, whole-image residual standard deviation lands between 0.08 and 0.44 in CIELAB lightness units. In one paired regeneration example, reference cleaning reduces output debris density by 45%.

The abstract closes with a jab at the field: 'Mi-Ripple links measurable artifact reduction to visibly cleaner generated images, rather than optimizing a spectral score alone.' No baseline systems are named, and the abstract does not describe a broader dataset. It lands in a busy week of computer-vision papers on our tracker, one of only 15 AI photo stories we've logged in the last 90 days.