While the release of this person does not exist by three researchers created a massive wave of buzz, it never went beyond curiosity. When Ivan Braun and his team launched Generated Photos, the interest changed to concern and apprehension. Generating human portraits via AI was no longer a research paper but a real, viable business. One that can and probably will profoundly disrupt a multi-billion industry led by the likes of Adobe, Shutterstock and Getty Images.
While GAN generated images do not come as a surprise to readers of Kaptur – we wrote many times about it in the past – we couldn’t resist sitting down with Ivan Braun, co-founder of Generated Photos and learn more :
– A little about you, what is your background?
I’m a UX designer. I used to work on user experience in the early 2000s, then ran a UX agency. Recently, we switched to our own product, Icons8, a collection of resources and software for designers.
– What is Generated.Photos? What does it solve?
From the outside, it’s a collection of 100k photos. From the inside, it’s one of the many experiments we have. We create pictures and try to improve their variability. One of our other experiments is Photo Creator, a photo editor that allows for the creation of photos, not just searching for one.
We’ve got many experiments with generative technology. Most of them look like roadkill:
But not with Generated.photos which is the one we decided to make public.
It solves cost issues. If you ever organized a photo session, you know that you will have to deal with casting, styling, makeup, studio, lighting, and post-production. Each of these steps can cost hundreds or thousands of dollars, and at the end of the day, you have several dozen useable shots.
We, however, can create a hundred thousand images in a couple of hours.
Generated.photos can also solve racial, gender, body type, or ethnic bias. We recently interviewed an Indian team that can’t find the videos of middle-class Indians. Existing footage only shows poverty.
Why is that? To cover their costs, stock photographers must create images that will sell in volume and in doing so, feed into the current biases. To overcome this, you have to drop the creation price of a picture to almost zero.
Of course, we have still have a long way to go. We need to generate emotions, bodies, scenes, interactions. Then, repeat it all for videos.
– Right now, your database of images is on a google drive and hardly searchable. What’s next? A searchable database of pre-generated faces, or will you go for an AI that will generate faces on demand?
Right. Better quality, defects filtered out. Multiple filters that allow filtering people of a certain age, gender, etc., pretty much like on our other project with real people.
An on-demand generation is always an option too. Stay tuned for an AI model agency that we’re going to create.
– After faces, are you planning to offer more AI-generated content? What kind?
Different angles of faces, full bodies, dogs, cats, objects. Pretty much anything.
Then repeat it all for videos.
How far are we from full scenes stock photography generated by AI that can compete with current photos? ( like a couple walking on a beach or a family playing in a garden)
We’re on the way to do that. We have all the ground technology for that; all we need is a lot of patience and some GPU credits. Shout out to the cloud operators — sponsor us with hardware.
– Who is your typical customer, and what is she looking for when going to your website?
Speaking about Icons8, the parent company, they’re:
- Front-end devs
- Non-professional designers: school teachers, presenters, amateur videographers, and everybody who needs illustrations, icons, photos, and music occasionally.
– How do you measure success with generated images since it is free? Numbers of pictures downloaded? Overall traffic? Leads to your other sites?
By Leads. We’ve received interest for usages that we did not expect:
- From a simulation game for executives that simulates a factory blowup
- From self-driving car simulations. Apparently, most self-driving car training happens away from the streets.
- For anonymizing people. In a world of Google StreetView, people walk around with blurred faces. We can replace those with real fake faces.
And by the number of press mentions so far and partnership proposals. Finally, we won a contest.
– Beyond stock photography, where you see/predict other applications to your technology?
- Any creative photography
- Any visualizations such as cosmetic surgery
- Random people for gaming
- Simulations: for self-driving cars, for business games, etc.
Anything that could be created using generative technology:
- Knitted fashion
- Anything 3D-printed, etc.
– What do you answer people that say you are destroying the income of thousands of stock photographers ( and models)?
I feel for them because we’re stock photographers too.
Hey, stock photographers! There’s a new technology coming. Want it or not, it’s here. Let’s see how you can learn from it.
New technology in photography is nothing new. Celluloid film, digital photography, microstock — all caused disruption but also allowed millions of photographers who liked experimenting.
One of the ways to see it is you can get more from a single shot.
- You’ll be able to change the emotion, age, and appearance of your model with nothing more than moving a slider.
- You can change the position
- You can combine with any background
You can change the lighting. Full frontal flash? Good. Studio with $10k of softboxes? No problem, everything is a click away. What remains is your creativity. Choose, compose, and edit your scenes takes creativity. Your creativity.
– What would you like to see Generated Photos offer that technology cannot yet deliver?
Short term, I’d like to have a larger resolution. Generating is expensive, and improving resolution over 1024×1024 reaches the limits of the current hardware. We have to work around it.
In the longer term, more additive technology for custom production. With design being cheap, we all can have unique clothing available at the spot. I want to knit the sneakers in-store and sew the dresses in a fully automatic mode.
Author: Paul Melcher
Paul Melcher is the founder of Kaptur and Managing Director of Melcher System, a consultancy for visual technology firms. He is an entrepreneur, advisor, and consultant with a rich background in visual tech, content licensing, business strategy, and technology with more than 20 years experience in developing world-renowned photo-based companies with already two successful exits.