AI Case Study; Athens

Using AI To Generate A Night From Day Image for a Scene from The Plaka

April 2026; Production Dilemma & Solution

While visiting Greece in the spring of 2024, we were able to capture a day time image (see right) of the Plaka neighborhood in Athens. For a day|night backing, a photo-realistic night version needed to be created for print. Returning to the site to recapture original photography, both day and night, was impractical due to cost and length of travel, not to mention the scene had more then likely changed. Likewise, budget and deadline would not allow for the utilization of outside professionals in the creation of the night image and all work had to be performed in house, as is usually the case. The solution involved using a combination of adjustment layers, hand painting, night from day LUTs, and the use of AI to create the final night image.

Artistic Vision

As is the case with all creative processes, when using AI it is best to pre-visualize what you intend to create and then determine the means required to achieve that creation. Haphazardly generating imagery without starting points and clear direction is what produces slop in any medium, AI being no different.

Taking inspiration from Stephen H. Burum when lighting Prague, we asked ourselves if we had an unlimited lighting budget and weeks to work with, how would we light this scene at night? This image was captured in the early morning just after sunrise, but prior to the sunlight making its way into the alleyways, allowing for a long enough exposure to see into the first floor retail stores. More then likely, given the age of these buildings, those retail stores would be illuminated with fluorescent lamps, something that appears evident in the image. Enhancing this would make sense. Likewise, we can assume the above story apartments and the lobby on the right would be lit using some kind of warm tone light.

The early morning light added a slight gradient coming towards us in the alleyway on the left. Working with this, it would be easy to imagine having a daylight LED streetlamp just out of view producing a more intense gradient of light up the street. Likewise, this blue light, along with having the light above the retail transom daylight balanced too, added a great source of color contrast with the fluorescent and incandescent lights of the interiors. The holiday lights on the tree on the left could have been of any light temperature, but having them an intense orange provides more contrast not to mention the progression from blue to green to orange is a natural transition of colors.

Given it is not exactly clear what is behind the papered windows in the center, we worked on the assumption the space is a mezzanine of the retail store and felt fluorescent lights would make sense. Using artistic license on the pair of doorways in the middle, we created even more contrast by making one inviting and the other foreboding.

Creative Process

Starting with the original daytime photograph, we broke up the image into several overlapping vertical plates and processed each plate separately in Photoshop Firefly. As it currently stands, AI is limited to 4K output on resolution and working in this manner allowed us to generate a large enough night file for print. Furthermore, for aesthetic reasons, breaking the image into smaller portions of itself allowed us to further refine our artistic vision while also writing prompts easy enough for the AI engine to interrupt correctly.