In January last year, gaming startup Studio Atelico launched the Atelico AI Engine, which operates on-device rather than relying on costly cloud solutions. Six months later, the company secured $5 million in seed funding “to redefine the role of generative AI in games.”
Now, the team is developing its first title using its own tech. Bobium Brawlers, a turn-based creature battler, is serving as a testbed for Studio Atelico’s AI-first ecosystem, which uses generative AI trained on the studio’s own artists’ work. In this case, the AI is used to generate creatures.
Chief product officer Chris Olson tells GamesIndustry.biz that the studio knows that some will have concerns about Studio Atelico’s approach, particularly regarding AI ethics, and is “intensely aware and sympathetic towards some of the issues.” But he believes “attitudes can and will change” over time, particularly since the studio is training their AI model using their own artists’ work rather than public language models.
“We want the technology to kind of uplift humans. We’re not looking to reduce steps in the workflow, so you can fire half your art team. We have an art director. We pay our artists. We recently just released a document that’s in draft form, but it hints at where we want to go, called ARC, the Artist Rights Contract, and it’s so that an artist can use this as a way to interact and protect their livelihood and their art.
“It allows them to define what their art can be used for, gives them a chance to participate in the upside if their art is trained on the models, and very specific use cases. It’s only used for this game. You have to renegotiate if you’re going to use the art for other purposes.
“There are conversations that need to be had, and there are legal battles being fought, which they should be fought, but we think that you can start to define that path and create a fertile ground for everyone to benefit.”
Studio Atelico CEO Piero Molino backs up Olson’s point: “Because we train our own models for doing these things on our own art and our own, in this case, our own decks being created in-house, so it’s our own data. We are doing it ethically when we’re not replacing anyone here. We are just creating something that would not have been possible before with data that we own.”
On device
In a talk at Nordic Game earlier this year, Studio Atelico co-founder and game design director Ennio De Nucci detailed the studio’s technology, which is designed to run on-device rather than connecting to power-hungry data centres in the cloud.
“It plugs in as a library of code, it’s really easy to integrate, and you’re able to create whatever kind of game you want,” he said.
De Nucci expounded the benefits of on-device AI: namely, low power usage and the lack of a need for constant online connection. Plus, he pointed out that vast amounts of training data simply aren’t needed when you only want the model to do one thing, like character creation.
“Using frontier models is a complete overkill,” he said. “The training that we’ve done in our models to give us the kind of art style that we want for mobile products is not something that you can have with a cloud model.”
Most importantly, using on-device AI means developers don’t have to pay hefty fees to companies like Google and OpenAI for their generative AI models.
“The cost problem on the cloud still has not been solved,” he said. “It’s never going to be zero because you need to pay this provider that needs to make money out of this thing.”
Only with AI
Molino thinks that players will become “more attached” to creatures in Bobium Brawlers than in franchises like Pokémon, because the creatures come to life through personalised prompts.
“The unique thing about this game is the creature creation aspect that is only made possible thanks to the technology that we are developing,” he says.
“You’re creating something out of your mind,” he continues. “We’re establishing a much tighter relationship with the creative flow of someone. We found people giggling when something they’d typed popped up in front of their eyes. Yes, people have their favourite Pokémon because they have a decades-long relationship with them, but we think that we’re establishing a different type of special connection.”
Studio Atelico aims to use AI “to expand what’s possible,” rather than retrofitting AI into existing game concepts.
“We had about ten prototypes and game mechanics from that experimentation,” said Molino. “We discovered it’s like a spectrum. On one end, you have traditional game design. On the other, the open-endedness AI offers.”
“With Bobium Brawlers, we are hitting something that is somewhere in the middle, and our observation has been that the most fun comes with blending the two – classic game design with the new possibilities of AI. We wanted to make a game that was fun, but also relatively small in scope, because nobody’s figured out what’s a good recipe for an AI-based game.”
Scaling the vision
Studio Atelico’s ambitions have significant financial backing thanks to the $5 million in funding led by Air Street Capital’s Nathan Benaich, with participation from key figures at Hugging Face and Snorkel AI.
Olson is mindful of the high costs associated with user acquisition, but he believes the uniqueness of Bobium Brawlers and the studio’s AI engine will help them “figure out who we think the audience is.”
“Because it’s kind of fresh and new, and I think definitely we’re going to find some interested parties and hope we get some virality either on the school grounds or on a YouTuber influencer, that kind of action,” he says.
“We are not making yet another Match 3 game where you need to spend all your money in user acquisition because you have no differentiation fundamentally,” Molino explains. “We have a differentiation, and we believe that is going to be the main driver of everything.”
Molino positions the game as a “blue ocean” opportunity within a stagnant mobile market. Unlike AI-integrated games that utilise subscription models to offset cloud costs, Atelico’s on-device engine has allowed them to launch Bobium Brawlers as a free-to-play title.
But ultimately, Bobium Brawlers is intended as a showcase for the Atelico AI Engine. The goal is to transition the engine from a robust, “leaky” architecture to a toolbox that allows developers to blend AI components as they see fit, rather than relying on fixed cloud APIs.
“Our long-term aim would be to be the Epic Games of AI,” Molino says. “Where you both do the games that are made possible by new technology, and then the technology helps create new games. It’s a self-reinforcing positive loop.”
The studio plans to develop small games, gather feedback, and iterate on future projects. “Depending on each game’s success, we aim to release about one game per year, allowing us to learn what works and refine our approach based on feedback.”
“Our long-term aim would be to be the Epic Games of AI”
Kiero Molino
Molino compares this iteration roadmap to Doom, which was Id Software’s fourth first-person shooter, but its biggest success.
“They put out HoverTank, and nobody cared for it. They put out Catacombs; nobody cared for that. They put out Wolfenstein, and people started to like it, but Wolfenstein then became more successful after Doom was released, because people then went back, ‘Oh, who are these guys?’ and then figured out that they had made other shooters before.
“They iterated quickly because they had a new tech, nobody had figured out yet how to make a first-person shooter – it was a new concept, so that iteration was needed to get to the point where they had a polished product that people actually loved. I believe that we will be in a similar situation.”
