Real? Fake? Does it matter?

From photo hunting to the backdrop of my dreams – how AI simplifies my process

It’s August 18, 2021. I’m sitting at my desk, scrolling through various stock image sites in search of the perfect background image for a photorealistic composition. To be more precise: a vehicle needs to be staged in its natural environment – the road. The client who commissioned our agency to do this wanted to save money. A professional photo shoot in the middle of the city would be too expensive. Useful 3D data of the product? Not this time. Yet it would reduce the required effort considerably: vehicle, surroundings, lighting mood, reflections, depth of field – all of these could be defined and controlled. What remains is a modest selection of retouched cut-outs from different perspectives. Doesn’t sound bad at first. But now the hours of research for a professional background image begin: the right angle, a suitable setting (an old town street in Germany), landscape format. Not so easy – but not impossible either. Less than a week later, a convincing composite image is redy – real-looking, high-quality, flexible for print and online use.

Four years later.

Today, my workday looks very different. The hours I used to spend researching now go into the precise creation of AI-generated backgrounds. Like a digital painter, I can design every detail the way I (and my customer) envision it.

In concrete terms, this means:

Nowadays, if I want to integrate a product into a realistic scene, I first analyze the perspective of the cut-out. In this way, I avoid disproportionate compositions. Then I define the surroundings – whether it’s a city center, an alpine country road or an orbital highway on Saturn. Anything is possible. And everything can be generated.

And what about AI humans?

In short: pretty good. Why?

A few months ago, we developed a campaign to promote a digital customer portal. The plan was to use international testimonials – real people, real users. The idea: people from different countries would appear as testimonials. A professional shoot on location would make sense. But then we got the news: the testimonials had neither the time nor the inclination for elaborate photo productions. A problem? In the past, yes. Today, no. Because we had already worked intensively on the AI-supported creation of images of people.

Our problem:

Personal photos of the testimonials were not an option – sensitive in terms of data protection, but understandable from a human point of view. Who would give vacation selfies to an agency?

Our solution:

A smartphone, a white wall – and a little dance.

As the scenes were only supposed to show everyday situations anyway (people at the computer, walking or sitting), 20 simple cell phone photos were sufficient. The testimonials turned slightly on their own axis, changed facial expressions – and the character reference was ready. No stress, no private pictures.

We used this to train the AI. The desired scenario was then described – and if necessary, we retouched small details such as birthmarks or scars. For the finishing touches, we added real background images and elements from the 3D artist, such as monitors or keyboards. The latter are still a weak point of many AI systems today.