Skip to content



AI likes hot people. Including Santa.

By now you’ve probably noticed that most of the people you see in AI-generated images are unusually attractive. Smooth skin. Brilliant smiles. Eyes that pierce and hair that tumbles in lustrous waves.

Even when the prompt specifies nothing in the way of appearance. Even when the prompt specifically requests an “ugly” person.

Maybe it’s because AI models are trained on images that represent our best versions of ourselves – styled shots of twentysomething It Girls and famous himbos. One popular AI training data set is called CelebA, consisting of more than 200,000 photos of celebrities, according to an article in the Atlantic. 

Or we could point the finger at… ourselves. Tools like Midjourney and DALL-E are also trained on civilian selfies, which we tend to post because we feel we look our most attractive in those moments. 

AI learns from and responds to the images we prefer to interact with. For example, Adobe's Firefly uses data on which images we download the most to inform its model, which can create a feedback loop that establishes attractiveness as a standard. 

We humans love an average face. Not average as in “so-so” but in the literal sense: a midway point between extremes. “Averageness is more attractive in general than non-averageness,” Lisa DeBruine, professor at the University of Glasgow School of Psychology and Neuroscience, told the Atlantic. AI, too, prefers average, it turns out. As a result of the way they process data, AI image generators tend to combine faces, creating their own that are more symmetrical and blemish-free. 

Which is a very long way of explaining why the AI-generated Santas you see here look like models in some holiday fever dream.

Who knows, maybe Santa is a smokeshow. He exists only in our collective consciousness, which itself has been trained on books, movies, and advertisements.

Happy holidays from all of us at R/GA. Let’s keep it real… whatever that is.


Tools used

GenAI Art Direction

Content at Scale

Using Midjourney, we designed a series of images with prompts focused on specific photographic styles and locations. For greater efficiency, we used batch scripting to produce a large (1000+) quantity of visuals, fast. We manually made selects, prioritizing images that we felt brought our narrative to life and struck a balance between “plausibly hot” and “outlandishly hot” (e.g. too much smoldering eye contact, too many abs). For Instagram Stories we animated Santa singing an original song titled “Santa Is a Smokeshow Now,” lyrics by ChatGPT. (Prompt: “Write a holiday song titled ‘Santa Is a Smokeshow Now’ that is about how Santa got really hot.”) Music by Suno. Finally, we put together everything in Figma and handed it over to our dev team for that final bit of holiday magic.

Unique cards
Singing Video
Generation Time
Image Styles
Discard Rate