Larian will no longer use GenAI for Divinity concept art, and any genAI used for other games will be "trained on data we own"

Larian CEO Swen Vincke has provided an update as to how the studio are planning to go about handling the use of generative AI in the making of Divinity. In an effort to alleviate concerns, the newly decided plan is to avoid using it in any capacity as part of the concept art development process. In general, Larian are aiming to ensure any in-game generated assets in their games come from models “trained” on data Larian already own, rather than risking the charge of creative theft or actual copyright breach.

Vincke revealed these plans in an answer to a question about the studio’s Generative AI use submitted to the public AMA session the studio have hosted today, January 9th. The exec kicked off by stating that “there is not going to be any GenAI art in Divinity”, though he seemingly isn’t ruling it out for future games.

“I know there’s been a lot of discussion about us using AI tools as part of concept art exploration,” he continued. “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.”

Further on in his response, he outlined the studio’s current plan when it comes to using generative AI to make art in games at large. “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,” Vincke wrote. “If we use a GenAI model to create in-game assets, then it’ll be trained on data we own.”

Comment
byu/Wombat_Medic from discussion
inGames

To see this content please enable targeting cookies.

Manage cookie settings

In between those two points, the CEO made clear that Larian still plan to experiment with generative AI in other areas. “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,” he wrote. “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.”

Elsewhere in the thread, Larian’s machine learning director Gabriel Bosque offered an explanation of an area in which the studio are making use of this sort of tech. “There is currently one example of machine learning generated assets that end up in the game and that is within our cinematics and animation pipeline,” he wrote. “In this pipeline we try to capture the actor’s performances as best as we can, so we use machine learning models to clean, retarget and even add motion when it’s not motion captured. These models are trained exclusively with Larian data.”

In my eyes, the changes outlined by Vincke are generally positive, though the matter of controlling and knowing definitively where every bit of data a model has pulled from originates feels like something which could be tough to pull off in practice. I’m no expert, though – perhaps the solution is as simple as locking your herd of generative AI models in an isolated room that only contains internal data. In the long run, there’s also the question of whether ‘acceptable’ usage of generative AI like this might lead to more wholesale and unregulated adoption down the road.

For now, at least there’ll be no weird AI portraits in whichever dungeons Divinity sends you to investigate.