Painting with Content: AI Will Accelerate Creative Work, Artomatix CTO Says

by Brian Caulfield

Video games have evolved into sprawling virtual worlds that can take months to explore. Retailers are building ever more detailed 3D environments to showcase their wares. And in a high-definition world, digital effects in movies and television grow ever more elaborate.

Demand for quality digital content is exploding, and so are the costs involved in creating it.

The solution, explained Artomatix CTO Eric Risser to scores of technologists, designers and entrepreneurs crowding into a talk at GTC Europe Thursday, will be to harness deep learning to combine the speed of computers with the flexibility of human artists.

Painting with Content

“In the future artists aren’t going to paint with colors anymore, they’re going to start painting with content,” Risser said. “They’ll quickly sketch what they want, then they’ll have the AI go and fill in the content.”

One of many breakthroughs, Risser said, came in 2015, when a paper authored by Leon A. Gatys, Alexander S. Ecker and Matthias Bethge — building on decades of research into an obscure field called “texture synthesis” — detailed how to use convolutional neural networks to quickly turn a small sample of a texture into patterns that appear natural to the human eye.

That work sparked an explosion of innovation, with neural networks being built into sophisticated production tools for video games and movies offered by shops such as Dublin-based Artomatix.

But video games — where a sprawling open world game can take five years and $250 million to create — are just the start. Deep learning promises to help humans create everything from architecture to fashion to movies and, ultimately, rich virtual worlds more quickly and easily, Risser said.

Based in a former Guinness brewery, Artomatix uses machine learning and big data analytics to produce everything from detailed landscapes to armies of bad guys, each different from the next. The company won our $100,000 Early Stage Challenge at NVIDIA’s Emerging Companies Summit in 2015.

 The Zombie Graph

Its work hinges on a phenomenon explained by what Risser calls a “zombie graph.” When you create an example of a video game monster, explained Risser, who holds a Ph.D. from Trinity College Dublin, it contains a great deal of unique information. Add another zombie, and you get a little more unique information.

After a while a pattern starts to form, with the amount of unique information added with each additional zombie decreasing, even as the number of examples grows.

Artomatix uses machine learning to make the most of the unique information offered by just a handful of examples of zombies to generate an entire army of unique-looking ones.

That saves artists time — and drudgery — while giving gamers the variety they need to immerse themselves in a digital world. It’s an example of a trend that promises to change the economics of content creation, for the better.

And just in time.

Everyone from architects to auto designers to furniture companies are building three-dimensional models to design and sell their goods, Risser said.
And while humans, who Risser describes as “creative nuclear reactors,” can create this content, there’s just much more work than there are people.

“That’s where creative AI is really disruptive, because it’s been able to break through this supply ceiling,” Risser said.

Content for the Next Generation of Social Media

Meanwhile, immersive, collaborative VR technology experiences, like NVIDIA Holodeck, will morph from a tool for designing real-world products into environments where richly detailed virtual worlds can be enjoyed.

“That’s going to be the next generation of social media,” Risser said. “You’re going to build a custom space, unique to you, and you’re going to want to fill it with unique stuff that is uniquely yours — this is where creative AI can come in and build virtual worlds for you.”