As photographers wait nervously for artificially intelligent (AI) robots to replace them, an inherent quirk of the technology may actually be what saves them.
For programs like Midjourney and other AI image generators to produce good results they need quality source photos to begin with.
Recently I’ve been playing around with Photo AI, a program that claims to be the “world’s first AI photographer”. To build a model of someone, you have to feed the program photos of that person and I quickly realized the better the photos are; the better the results you get from Photo AI.
In a recent interview with PetaPixel, photographer Sophie Gammand talked about how she gave rescue dogs their ears back with the help of AI and she made clear that the pictures worked so much better when she used her own high-quality photographs.
“The photo needed to be clear and the angle easy to work with,” she explains. “When I used people’s own photos, I struggled a bit more than when I used my studio portraits.”
That’s because photos taken by lay people tend to be of far lower quality when compared to an accomplished photographer such as Gammand — something that people working on AI image generators must be aware of.
How Will AI Learn New Objects and New People Without Photographers?
Say someone wants to create AI images of their new product. Well, you need source photos. But what if that person doesn’t understand how to get a decent photo of their product in the first place? We all know someone who can’t take a photo to save their life.
Google recently launched a new product photography tool aimed at helping businesses to create their own high-quality images without the need for an “expensive” photographer. In the promotional material, it shows really clear photos for the AI to work with — perhaps ones, ironically, that were taken by a pro photographer. But will it work with terrible source photos taken by people who have no understanding of how to manipulate light, frame a photo, or even hold a camera still? I doubt it.
AI photos are dependent on real photos. The stunning images coming out of Midjourney that look like real photos exist only because there are already billions of amazing photos that look like that which the company used without the photographers’ permissions.
If people or clients or whoever want bespoke high-quality AI images that are unique, well first you need high-quality, unique photos.
Ultimately, there are things in the future that the AI will have no reference for. It will need a source and that source has to be a set of high-quality photographs taken by a trained photographer who knows what they are doing.
Personally, I see the AI and photography spaces overlapping with AI technicians relying on the work of photographers who can take high-quality images which can be used to train a model.
Better yet, photographers could become AI technicians (surely it’s not that different from photo editing work) creating more markets and economic opportunities for the talented men and women who have honed their camera skills. I certainly hope so.