Thoughts on the evolution of AI into cameras. Using AI to automate camera operations and image capture.

There are two main phases in this process, and right now, we’re at the turning point between them.

Phase one, where AI acts as a better technician, is basically complete. Features like autofocus with subject, animal, or eye tracking, scene-based exposure, real-time noise reduction, and multi-frame computational stacking are now standard on mid-range and high-end cameras. Canon, Sony, and Nikon’s deep-learning AF systems are good examples. Here, AI automates the technical side of taking photos, which is clearly a positive step. It makes it easier to get the shot you want without changing the nature of photography.

Phase two, where AI becomes a co-author of the image, is where things get more interesting and complicated. Now, the latest developments involve generative reconstruction inside the camera’s processing: relighting, creating skies or details, and denoising that invents realistic textures instead of just recovering them, all happening on the device in real time. At this point, the line between capturing and generating an image starts to blur. A photo is no longer just a record of light hitting a sensor; it’s increasingly shaped by a model’s idea of what a good photo of that scene should look like.

This change is why the industry introduced a countermeasure this year: C2PA hardware signing, which is now available on Sony’s professional cameras (Alpha 1 II, Alpha 9 III, FX3/FX30) and is being added to Canon’s Authenticity Imaging System for newsrooms. It’s a meaningful step, but there’s an important limitation to note. C2PA can prove where and when a file was created (for example, this file came from this sensor at this time), but it cannot prove that AI generation was not involved. It only records what the camera or creator claims, not whether the image was synthesized. Nikon’s attempt actually had problems—they released C2PA on the Z6 III, discovered a security flaw, and had to revoke all certificates, which still haven’t been restored. This shows that the “trust infrastructure” is still not as mature as the generative technology it’s supposed to oversee.

Here’s my real opinion: mechanical automation is a clear benefit for working photographers because it lets them focus on composition and timing instead of struggling with their equipment. I’m more cautious about the generative side, not because the technology is bad, but because it quietly shifts who is really creating the image. When a camera decides what a “good” sky or face should look like, the photographer’s job moves from capturing reality to choosing from the AI’s version of it, often without making that clear to the viewer. In art and photography, this has two effects. It makes things more accessible, since technical barriers keep dropping, but it also makes results more similar, as everyone’s “AI-optimized” photos start to look the same. The systems for proving authenticity, like C2PA and whatever comes next, are going to become much more important than most people realize, especially since the technology is moving faster than these safeguards.