The marketing name is "night mode"; the filing name is hybrid low-light image enhancement. Motorola's granted patent US12079973B2, issued September 3, 2024, claims a hybrid method — explicitly mixing classical and neural processing — to enhance low-light images. Its CPC tags G06T 5/92 and G06T 3/4053 are dynamic-range-enhancement and image-scaling classes.
On the record, night mode is one of the more impressive feats of computational photography because it appears to defy the hardware. A small phone sensor in near-darkness captures very little light; the bright, clean night-mode result is reconstructed from a burst of noisy frames through alignment, fusion, and now neural enhancement. The "hybrid" framing is honest — it is not pure AI, it is classical multi-frame processing with neural steps layered in.
Why the hybrid detail matters: a purely neural approach can hallucinate detail that was never there, while a purely classical approach hits a noise floor it cannot beat. Combining them — classical fusion to recover real signal, neural enhancement to clean and sharpen — is the pragmatic engineering answer, and claiming a specific hybrid is more defensible and more honest than claiming "AI night mode."
Novel, or just renamed? Low-light enhancement is well-trodden art; the claim's interest is the particular hybrid combination on-device. As always, the principle is established; the specific method within a phone's compute budget is the work.
The strategic frame is that low-light performance is one of the last camera battlegrounds where buyers can see the difference, which keeps it a place worth patenting. Motorola, not a sensor leader, competing here through processing IP shows the differentiation has moved decisively into software.
Follow the filing, not the sample photo. A night-mode shot that looks impossible for the hardware is exactly that — reconstructed, not captured. The 2024 grant names the method: a hybrid of classical and neural enhancement, dated and classified.