The tool remains, but its use is more limited
It goes without saying that the subject of AI was going to be the first discussed during this question-and-answer session around divinity. Swen Vincke therefore responded to the concerns by initially stating that there may have been confusion, while affirming that he will no longer use generative AI to create concept art :
“ I know there has been a lot of discussion around using AI tools for the concept art phase. We have already clarified that this does not mean that these concept arts themselves are generated by an AI, but we understand the confusion this has caused. In order to remove any ambiguity, we have decided to no longer use generative AI tools when developing concept arts. In this way, the origin of the illustrations cannot be the subject of any discussion. »
Larian reaffirms that there is no question of resorting to the use of generative AI to produce artistic elements in the game, and the same goes for the writing. Adam Smith, in charge of writing, explains that no lines of dialogue will be generated this way :
“ We had a small group testing the tools for generating text, but the results were 3/10 at best and these tools were only there for research, not for use in Divinity. Even my worst writing – and there are A LOT – are at least 4/10 (although Swen might disagree :p), and the number of iterations it takes to have a quality line of dialogue is enormous. »
The CEO nevertheless adds that the studio necessarily observes advances in AI so that this tool helps production go faster in certain non-creative sectors :
“ That said, we are constantly looking to accelerate our experimentation processes. The more iterations we do, the better the gameplay. We believe that generative AI can help us with this and so we are conducting trials in all departments. We hope it will allow us to refine our ideas more quickly, leading to a more focused development cycle, less waste, and ultimately a higher quality game. It is important to note that we will not generate any creative for the game without being absolutely certain of the origin of the training data and obtaining consent from the people who created it. If we use a GenAI model to create assets for the game, it will be trained on data that we own. »
The tool is therefore not completely thrown into the trash, but the studio at least promises to use it with a little more care.