There are often performance factors people underestimate. For example Portrait Lighting can run on hardware without the neural engine, if you hack the exif to

There are often performance factors people underestimate. For example Portrait Lighting can run on hardware without the neural engine, if you hack the exif to say it was shot on a newer device. However, this is just when editing. There's no way it would be capable of the realtime preview, and casual users rely on the preview to make the feature useful at all. Actually, it would confuse a ton of users. We added portrait-mode support for non-humans on the iPhone XR. The hardware doesn't provide realtime depth information, so we can only apply the effect after you've taken the photo. We frequently receive support requests telling us that the feature doesn't work. I can't speak to why Deep Fusion isn't coming to old devices, since it's an offline process. I mean, I could speculate that if it relies on multi-cam capture to capture multiple images, that's strenuous API that isn't supported on hardware earlier than the XS. The cynical perspective is that Apple caps features to the latest hardware to sell new hardware. I'd say it's just as likely that Apple looks at the new hardware coming down the pipeline and plans features that push the hardware to its limit.