Larian's dual-track AI policy is a game development AI blueprint for risk-averse studios
Summary
Larian banned generative AI from all creative output in Divinity: art, writing, and dialogue
The studio simultaneously deploys proprietary ML trained on Larian-owned data for pipeline acceleration
Clair Obscur lost two Indie Game Awards over AI assets, setting the cost baseline for creative AI exposure
This split model is replicable: protect consumer-facing IP publicly, capture workflow efficiency privately
"There is not going to be any GenAI art in Divinity"
01
GenAI banned from all consumer-facing creative output
The ban covers concept art, writing, dialogue, and journal entries in Divinity
Writing Director Adam Smith confirmed no text generation tools touch any written content
Prior ambiguity around AI concept-art experiments forced this harder public line
"That way, there can be no discussion about the origin of the art"
02
Proprietary ML deployed quietly for pipeline acceleration
Larian is testing ML tools across departments to increase iteration speed
Any model used must train exclusively on Larian-owned data and assets
Vincke previously framed ML as automating "tasks nobody wants to do," not replacing creative roles
"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"
03
Industry backlash sets the cost baseline for AI exposure
Clair Obscur had two Indie Game Awards revoked, including Game of the Year, over AI assets
Assets were patched out post-launch, but awards were still stripped
Retroactive reputational damage is permanent and not mitigable after discovery
Smith rated GenAI text output "3/10 at best" after internal testing
"Even my worst first drafts are at least a 4/10, and the amount of iteration required to get even individual lines to the quality we want is enormous"
04
What studios and investors should watch
The Larian model is a replicable framework: hard public ban on creative AI, quiet proprietary ML for workflow
Studios lacking clean IP ownership of training data cannot safely replicate this approach
Data provenance is the critical diligence item for any studio claiming internal-only ML use
If award bodies or platform holders formalize AI disclosure rules, post-launch patching stops working as mitigation
Studios in pre-production should set AI governance policies before external perception hardens
Events
Companies
Games
Locations
—
COMPILED BY GAMES ATLAS EDITORIAL
RELATED ARTICLES
EXPLORE MORE
Comments (0)
No comments yet. Be the first to share your thoughts!