One of the first lessons of our exceptional commercial photography professor, the late János Rátki, was that studio sets are not built for eternity. Lesser educators would've gone for something pathetic, like "nobody cares what's outside of the frame of your photo", but what he meant was more along the lines of "feel free to duct tape that foam-core board to the broomstick and hang it off the ceiling lights to get the fill light you want".
A LinkedIn coach would say he's suggestion was solution oriented. To me it defined a more hands-on, doer attitude when doing visual productions both as an art director or a photographer ever since. An attitude that I'm hearing is less and less relevant in the age of AI. But as long as we mainly talk to AI in writing, I beg to differ.
And here is an example why.
I was toying with the idea of this watery, splashy shot for a while. A set notoriously hard to pull of, because the reflections of the water limit the camera angles and dictate the framing of the image a lot. A huge pool, a 10x10 scrim and a large studio of course gives way more breathing room, but I had to make due with much less.
A quick solution would've been to take the shot of the glass separately and unleash Adobe to change the background with generative AI. But how do you tell the nuances of random ripples and glints through a text prompt? How do you prompt reflections that are not visible on the "plate" shot? Well, you can't. You will end up with options that fit ok, but not ones that feel ok.
For me this was not the way. But to be more accurate I needed to find a common language with the AI tool. A common language that could convey the nuances that grounded the image in reality. And one that takes off the impossible task of figuring our non-existent reflections of the AI tool's shoulders. It was easier to do the prompting visually.
And visual prompting means doing.
So I started to build my little pool out of an IKEA tray and some cinefoil, set up home made scrims and small Godox Photo Equipment Co Ltd lights, flagged the reflections of the close-by walls, pushed my composition as wide as my set allowed, and took the shot. From here, I had my Rosetta stone to talk to AI. The task is still something I would've never done by hand. Something that needs actual image generation from nothing. But with enough reference for the mood, size and variability the final image looks and feels real and random at the same time.
Carving these Rosetta stones by building references for visual image generation is key at the current stage of generative AI tools and their prompting options. The space where the joy of hand crafting key elements meets smart hand-off to generative AI is a new, inspiring area, promising more opportunity to smaller productions, quicker delivery on the commercial market and an even broader offer for large-scale content generation.
Turns out if we use AI to be a doer, doers have an easier time talking its language.