BG3 studio Larian confirms it dropped genAI from Divinity

But doesn't rule out genAI assets in-game if it owns the training data

Larian Studios has taken the chance of a Reddit AMA to answer some of the most pressing questions from its fans concerning its upcoming game, Divinity, presented with a shocking trailer at The Games Awards 2025.

After the overwhelming acclaim earned through Baldur's Gate 3, a fantasy role-playing game set in the world of Dungeons & Dragons, it was hard to believe that the studio could do anything to shake the goodwill fans felt. However, Larian Studios' co-founder and game designer Swen Vincke's comments to Bloomberg on the studio's use of generative AI seemed to do just that.

Polygon reached out to Larian for clarification on the studio's generative AI usage, and Vincke provided a statement confirming that the studio was not releasing a game with "any AI components" and that they were not "looking at trimming down teams to replace them with AI." The full statement can be read here. The Reddit AMA further clarified their stance.

Answering a fan who asked what the studio's opinion was on the role of generative AI in the game development process, as well as its purpose for Larian in particular, Swen Vincke released the following statement.

So first off - there is not going to be any GenAI art in Divinity.

I know there’s been a lot of discussion about us using AI tools as part of concept art exploration. We already said this doesn’t mean the actual concept art is generated by AI, but we understand it created confusion. So, to ensure there is no room for doubt, we’ve decided to refrain from using genAI tools during concept art development. That way there can be no discussion about the origin of the art.

Having said that, we continuously try to improve the speed with which we can try things out. The more iterations we can do, the better in general the gameplay is. We think GenAI can help with this, and so we’re trying things out across departments. Our hope is that it can aid us to refine ideas faster, leading to a more focused development cycle, less waste, and ultimately, a higher-quality game.

The important bit to note is that we will not generate “creative assets” that end up in a game without being 100% sure about the origins of the training data and the consent of those who created the data. If we use a GenAI model to create in-game assets, then it’ll be trained on data we own.

The statement is a substantial change from Larian's previous position on the subject, which leaned toward using genAI models to help "flesh out PowerPoint presentations, develop concept art and write placeholder text." According to Vincke's answer here, however, that's no longer the case.

It doesn't rule out the use of generative AI altogether, however, as Vincke is making an important distinction: Larian believes that using genAI behind the scenes can help the creative process and will continue to tinker with it, but nothing made with genAI will end up in the game unless they own the training data or have the consent of the owners of the training data.

Larian's machine learning director, Gabriel Bosque, went on further to clarify how the studio's use of machine learning tools influences "game development pipelines of the future" as well as creative assets. Answering a fan's question on how AI is being implemented and how it makes game development smoother, Bosque answered with the following.

"This is all new technology, so I totally understand why it’s difficult to see where the positive uses are. We believe Machine Learning is a powerful tool to accelerate and make game development more efficient and streamlined. This means that our creatives have more time doing the work that makes their jobs more rewarding and the game a richer experience.

The important bit to note is that we do not generate “creative assets” that end up in a game without being 100% sure about the origins of the training data and the consent of those who created the data.

With over 100,000 voice lines scheduled to be in the game, recorded by hundreds of actors, and even hundred thousands more to be recorded that will not end up in the game, there are useful tools for us to reduce the ‘mechanical legwork’ and free up time for teams that would be bogged down doing tasks that kept them away from what they really want to be doing. Additionally, our ML R&D team monitors and experiments with anything that is state of the art and that might influence game development pipelines of the future. This is important to us because we make our own engine.

But we draw lines in the sand too. We explicitly committed in our actor agreements to not using the recordings to train or build AI voice modelers, because we are aware of how sensitive it can be to artificially generate an actor’s voice. Even if an actor were willing to agree to this if we'd compensate them, we don’t currently feel comfortable with including an AI-generated voice in our games."

The use of AI to steal from actors has long been an unfortunate occurrence within the games industry. Larian's position here suggests that the studio has no intention of using AI to generate actors' voices, which will be a comfort to actors in the industry.

It's still up for debate whether Larian's answers here will satisfy the company's fans, but at least the company has taken the time to address the complaints, clarify its position, and even take a step back on the topic of genAI in concept art.