Larian says no AI for concept art or writing - here's what that means for writers
Larian Studios drew a clear line in a Reddit AMA: no generative AI for concept art or writing on its next game, Divinity. This comes after criticism sparked by earlier comments about experimenting with AI for idea exploration and placeholder text.
CEO Swen Vincke put it plainly: "There is not going to be any GenAI art in Divinity." He added, "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."
On text, the studio also clarified it won't use gen AI to write the game's words. Early "placeholder" tests aren't the plan anymore. The team emphasized human authorship across narrative and dialogue.
There's one caveat: Larian left the door open to use AI for future in-game assets only if the model is trained on data the studio owns. That's a signal about control, provenance, and risk management-not a shortcut to content at scale.
Why this matters to writers
Readers and players care who wrote the words. Clear authorship builds trust; muddled provenance erodes it. Larian's stance shows that audiences reward teams who can point to a human voice behind the work.
Experimentation is fine, but boundaries matter. If AI touches anything that ships, be ready to show the chain of custody: who wrote it, who edited it, and what tools were used. That's where many teams stumble.
Legal risk also enters the picture. Training on owned data reduces exposure and makes provenance straightforward. If you need a primer on current guidance, see the U.S. Copyright Office's resources on AI-generated material: copyright.gov/ai.
Practical takeaways for writing teams
- Draw a visible line. Publish a one-sentence policy: "No AI writes or rewrites shipped narrative, dialogue, or marketing copy." Internally, you can still test AI for outlines or research-but label it clearly.
- Define tool boundaries. Allowed: brainstorming prompts, beat sheets, structure checks. Not allowed: drafting story beats, character voice, dialogue, or lore that will ship.
- Track provenance. Keep a lightweight log: who authored the draft, what edits were made, and whether AI assisted. Store prompts and outputs used for ideation in a separate folder.
- Protect your voice. Build a living style guide and character bible. If you use AI at all, use it as a critic (find plot holes, flag contradictions), not an author.
- Set review gates. Human editor sign-off is non-negotiable. Run checks for style consistency, factual accuracy, and IP continuity before anything leaves the building.
- Own your data. If you ever fine-tune models, feed them only with content you control and license. Keep training corpora documented for future audits.
- Be public about it. A short, plain-language AI policy calms readers, aligns the team, and saves you hours of back-and-forth online.
What to watch next
Larian's approach is a template: human-led writing, clear provenance, cautious tooling. Expect more studios and publishers to formalize similar policies as legal and audience pressures tighten.
If they do adopt AI for non-narrative assets later, they've already set the standard: use owned data, clarify scope, document everything. Writers can adopt the same posture without slowing down the work.
Resources for writers building ethical AI workflows
Curated tools that support writing without crossing authorship lines: AI tools for copywriting (curated). If you're organizing team upskilling, see role-based options here: Courses by job.
Your membership also unlocks:
