AI; The Future is Here Whether You Like It Or Not!

[JoeKitchen]

I have no idea where to put this, so I feel like this would be the best place. 

Given what I have seen in the last few months, I honestly feel in years to come March 2026 will be regarded as the definitive paradigm shift with AI.  In February the tools were still not ready for prime time.  Now though, they most certainly are. 

Attached are two day time panos I captured with a Phase One back, each around 300 MP.  For my backdrop business, we often need matching night versions to print a day|night backing.  For paid gigs, we stay all day to get both.  For stock, I typically am working more for quantity then spending an entire day for one view.  So for most stock images I only capture the day version and, if a production needs a night version for the drop, would either return to the site or spend the 20+ hours painting it in in post.  These night versions were generated with AI in about 90 minutes each at the same resolution as the originals. 

Honestly, the hardest part was figuring out what prompts to use to ensure nothing changed.  Then it was a matter of playing around with it to get what I wanted. 

Right now, we could send both pairs to print for two day|night backings up to 30 feet tall. 

Sure, when pixel peeping at 200% they are not as good, but they do not fall apart either.  And of course the actual night images would have more nuance with variations in luminosity and color temps, and I probably left the first too warm, but holy shit!

I have my reservations on this tech, but the courts are moving too slow to stop it and politically it is a hot potato.  At this point, it seems either adopt or die on the vine. 

[Chris Kern]

Attached are two day time panos I captured with a Phase One back, each around 300 MP.  For my backdrop business, we often need matching night versions to print a day|night backing. . . .  These night versions were generated with AI in about 90 minutes each at the same resolution as the originals.

Quite remarkable!  Thanks for posting these examples.

Honestly, the hardest part was figuring out what prompts to use to ensure nothing changed.  Then it was a matter of playing around with it to get what I wanted. 

Yes, learning how to draft a prompt is an essential skill for the effective use of these generative AI tools.  My experience is that it's a good idea not to try to be too terse.  Qualifying phrasing, even to the point of redundancy, can help the neural network "understand" the nuances of a request, and often reduce the need to issue supplementary prompts (e.g., corrections) to fine-tune the initial response.  I wouldn't be surprised to see a market emerge for tech-savvy prompt writers who can "translate" what users want into efficient prompts.

Of course, if the demand became great enough, the human translators no doubt would quickly be supplanted by generative AI models that have been specifically crafted to "talk" to neural networks.

[Rob C]

Nightmare.

Truth to tell, we were party to our own decline from way back when we all thought stock was a great idea. You could say we innocently (?) cultivated our own first cancer cell.

[JoeKitchen]

Decided to go a lot more nuanced with these, including specifying color temps for individual lights.  If a production wanted to use them, these would still require about 1 to 2 hours of additional post to bring them to print readiness, but looking good enough for our catalog.  (Note: stars in night skies dont really work in backings, they end up looking like pinholes in the fabric and film so out of focus they would not be visible in any takes anyway, just in case you are wondering why there are none.)

Insofar as Rob's comment, I am looking at this from the perspective of Pandora's Box. 

Everyone knows that when Pandora opened the box, all the evil things inside quickly flew out into the world.  What most people don't know about the myth is that, in a scarred frenzy, Pandora quickly shut the box locking in all of the good that had yet to fly out. 

The concerns over AI are real, but AI ain't going back in the box.  So we might as well learn the good things it can do. 

[hubell]

Fascinating post. I am curious as to which AI program and prompts you used.

[kers]

interesting how well AI does this; it also invents the entire interior of the rooms you cannot see in daylight.
The incredible detail of the large format camera makes it look even more super real.
At the same time you use a camera to expose every real detail and use AI to fake every detail.
There is something odd about that.

[Slobodan Blagojevic]

Following.

[JoeKitchen]

I have been using Gemini 3.1 through Adobe Firefly.  This seems to work the best.  Plus with my pro account, I am protected from any liability.

It keeps on getting better and better. 

For example, beginning of this year it could only produce an American bond on brick walls.  Now it can produce a flemish and British bond. 

I just finished a paid job where AI came in really handy.  The production had a neon sign made for the shoot, but it was positioned incorrectly, about 10 feet from where it needed to be.  Moving it in post was easy.  However it rained just before the shoot and the red light was reflecting off of the street through a metal fence.  I used AI to move the scallop of reflecting light and it worked perfect.  I also added in additional house lights through the trees. 

[JoeKitchen]

Just to show another example, about a month ago we started working with a major feature that needed a few backings.  The first one we worked on was from a location in Midtown where we had access to the 6th story roof and 3rd story apartment.  For a variety of reasons, we decided to work with the roof top view for the actual backing, but since I had access, I photographed them both.  Any opportunity to get an elevated day|night view is worth it, plus we now have two new images for the stock library. 

By the time we photographed the 3rd story night image, most of the store fronts were closed with the gates pulled down.  So I just used AI to alter the day version to look like night and blended in the first floor and sidewalk into the true night.  See below. 

I also used AI to light up some more windows in the 6th story view too, which I also attached. 

[digitaldog]

My two cents and an example from another site discussing AI:
https://www.photo.net/forums/topic/725016-photoshop-version-276/#findComment-6052652

As for non-Photo AI work, I'm using ChatGPT to do some rather complex work in Photoshop, like making perfectly constructed 16-bit gradients, and the scripts it makes take about a minute. Even if I knew how to write PS .jsx scripts (I do not), it would take quite a while to do so. I've built dozens of these scripts. One needed to convert Cd/m2 values into 16-bit values to build pixels; not for the mathematically inclined like me. Took ChatGPT about 1 minute to do so. The result was a perfect target as requested.

* Bummer: there is NO 16-bit numeric data entry in the Color Picker. But it can be done via a script. IF you know how to write one or have AI.

[Chris_Brown]

For 25 years I've photographed various plants (annuals, perennials, shrubs, etc.) for a client. Some shots were garden scenes which were planted in the spring and photographed a few months later, when the plants matured, filled in and blossomed. Most of the photography—75%—was the plants in consumer containers on white in a simple studio setup, with a second set of photos of the some of these plant/container combos taken on location to someone's home. I'd find locations in the various towns we worked in (SoCal, New Hampshire, Illinois, Michigan) pay the home owner and get a signed property release. Standard procedure.

Last fall, the company was purchased by an investment firm and they, as usual, asked where costs could be cut. The marketing V.P. asked me for ideas and I tried Adobe Firefly for the first time. After learning how to give the best prompt, I realized I'd probably never have to go on location again.

Below is an example of a container combo photographed on white, and the Firefly result after only one attempt.

[Rob C]

https://huntingtonwitherill.com/emusings

A photographer's take (2023) on AI.