Xiaomi 17 Ultra AI applies lunar crater detail to eclipse photos of the Sun

The Xiaomi 17 Ultra produced at least one eclipse photo where the Sun, appearing red-orange, was overlaid with craters, reliefs, and lines characteristic of lunar imagery. The publication Frandroid identified this as a failure of the AI-based enhancement that smartphone cameras apply when a telephoto lens is used at high zoom on the Moon. The system misidentified the eclipsed Sun as the Moon and applied its lunar detail reconstruction pipeline to the wrong celestial body.

This enhancement technique is not unique to Xiaomi. The article references a prior Samsung controversy around its Space Zoom mode, which similarly uses AI trained on Moon images to fabricate crater detail that the camera sensor cannot optically resolve at x100 or x120 zoom levels. The Xiaomi failure simply makes the mechanism unmistakable: the AI does not capture what the lens sees, it replaces it with a learned reconstruction, and in this case it triggered on the wrong target.

Multiple users on the social network X reported the same behavior, sharing photos where the Xiaomi 17 Ultra turned the solar eclipse into a Moon-like image. This corroboration indicates the misfire is reproducible rather than a one-off hardware fault. However, Frandroid notes that not every eclipse photo from the device exhibited the problem, which implies the AI pipeline activates only under certain zoom or scene-detection conditions that were met in some shots but not others.

For engineers building computational photography systems, the failure mode is a textbook example of an ML model applied outside its training distribution. The lunar-enhancement model was trained on Moon imagery and apparently lacks a guardrail to distinguish the Sun, even an eclipsed Sun, from the Moon. The cost of this particular shortcut is credibility: when the reconstruction is wrong, it is visibly wrong, and it exposes the gap between optical capture and algorithmic fabrication in a way that is hard to dismiss.