Lead: Larian Studios, the developer behind Baldur’s Gate 3, confirmed this week that it is experimenting with generative AI in early development work for its newly announced RPG Divinity, while insisting final creative assets will be produced by humans. The comments by co‑founder and designer Swen Vincke followed the game’s reveal at The Game Awards and drew a mix of interest and backlash from fans and commentators. Larian says it employs machine learning tools for specific, non‑creative tasks — and that it is hiring more artists, writers and localizers rather than cutting staff. The studio also denies it will ship AI‑generated components in the released game.
Key takeaways
- Larian is using generative AI for ancillary tasks such as fleshing out presentations, generating concept outlines and producing placeholder text, not for final in‑game content.
- Co‑founder Swen Vincke reiterated the studio will not ship a game containing AI‑created components and is not planning to reduce creative headcount to replace people with AI.
- The studio states it currently employs 23 concept artists and is actively recruiting additional concept artists, writers, actors and translators.
- Vincke described three internal uses of ML: automation of tedious tasks, accelerating early white‑boxing iteration, and experimental work toward reactive gameplay — the latter remains exploratory and not production‑ready.
- The initial Bloomberg report and Larian’s comments prompted visible fan concern and conversation online about the role of generative AI in game creation.
- Larian frames AI/ML as a tool to increase creative time and improve iteration speed rather than a substitute for human creativity or craft.
Background
Larian Studios earned wide acclaim and commercial success with Baldur’s Gate 3, and its reveal of a new title, Divinity, at The Game Awards last week generated significant community attention. The timing placed the studio squarely in an industry‑wide debate about how generative AI should be incorporated into creative workflows, especially in art, writing and audio. Many developers and unions have recently raised concerns about automated tools displacing creative roles or enabling uncredited reuse of artists’ work. Larian’s response must be read against that broader backdrop: studios experimenting with tools while navigating player and creator expectations about authorship and labor.
Historically, game studios have adopted automation for tasks like animation retargeting, mocap cleanup and build pipelines to accelerate production without replacing creative decisions. Larian says it is expanding human teams—adding writers, concept artists and performers—while researching ML techniques to reduce repetitive work and improve iteration. The studio’s public posture follows conversations across games journalism and social platforms about transparency and safeguards when studios adopt new tooling.
Main event
The immediate thread began when Bloomberg reported Larian had “been pushing hard on generative AI,” prompting Vincke to clarify what that meant. He told Bloomberg the studio uses generative AI to draft PowerPoint content, generate concept art references and create placeholder text, but emphasized that final game assets will be written and produced by humans. After the Bloomberg piece and fan reaction, Larian provided a public statement underlining hiring activity and daily creative work by concept artists and writers.
Vincke expanded the explanation on social media, pushing back on any suggestion the studio is replacing artists. He said AI tools are used like other reference sources — to explore composition and ideas at early ideation — and that original concept art replaces any machine‑generated rough outlines. Larian reiterated that its goal with ML is to let people spend more time on creative work by automating time‑consuming, lower‑value tasks.
In an earlier interview with GameSpot, Vincke outlined three internal ML use cases: automation (mocap cleaning, voice edits, retargeting), white‑boxing to accelerate iteration, and exploratory research into AI‑driven reactivity for gameplay systems. He and the studio stress that the most sensitive creative domains—final visuals, authored writing and music—are not being divested to machine generation for released content, and that tools remain additive rather than substitutive.
Analysis & implications
At a practical level, Larian’s approach reflects a common compromise many studios are pursuing: adopt ML for repeatable, time‑consuming chores while keeping authorship and final creative choices in human hands. If implemented transparently, this can shorten iteration cycles, reduce QA burden and free artists to focus on higher‑value creative tasks. That said, transparency about which assets were assisted and how remains critical to maintain player trust and to address creator concerns about credit and reuse.
Labor implications depend on deployment details. Larian insists it is hiring and not trimming teams; the risk of displacement typically rises when studios prioritize cost reductions over creative investment. By publicly committing to expand concept art and writing staff while describing ML as a productivity tool, Larian lowers immediate displacement risk but does not remove future uncertainty if business priorities change.
From a design perspective, the most consequential long‑term possibility is AI‑driven reactivity — systems that generate unexpected permutations of narrative or NPC response. Vincke framed this as an experimental frontier for RPGs: ML could help detect narrative inconsistencies across hundreds of branching choices or enable emergent player interactions. These applications, however, require significant R&D and robust validation to avoid surface‑level or incoherent content being presented as authored narrative.
Comparison & data
| ML use case | Purpose | Production status |
|---|---|---|
| Automation | Mocap clean‑up, voice edits, retargeting | Active, production‑adjacent |
| White‑boxing | Rapid iteration of mechanics and layouts | Used in early design stages |
| Generative content | Concept references, placeholder text | Ideation only; replaced by human work |
The table above synthesizes Vincke’s descriptions: ML is applied where it reduces manual toil and speeds iteration, while creative authorship is retained for final assets. This distinction matters because automation and white‑boxing target productivity bottlenecks, whereas generative content for final release raises intellectual property and attribution issues. Larian’s statement that it employs 23 concept artists offers a concrete counterpoint to claims it is downsizing creative roles.
Reactions & quotes
After coverage spread, Larian released a formal statement highlighting recruitment and daily creative output by artists and writers, seeking to reassure stakeholders. The studio framed ML as “additive to a creative team” and said no AI components will ship in the game.
“We are neither releasing a game with any AI components, nor are we looking at trimming down teams to replace them with AI.”
Larian Studios (official statement)
On social media, Vincke pushed back directly against characterizations that the studio was replacing artists or aggressively automating concept art creation.
“We’re not ‘pushing hard’ for or replacing concept artists with AI.”
Swen Vincke (co‑founder, Larian) — social post
Media outlets framed the story as part of a broader industry conversation. Bloomberg’s reporting that Larian was exploring generative AI prompted much of the initial reaction and debate online, illustrating the sensitivity of the topic among fans and creators.
“[Larian] has been pushing hard on generative AI,”
Bloomberg (reporting summary)
Unconfirmed
- Whether any final in‑game assets will contain machine‑generated elements remains empirically verifiable only after release; Larian declares it will not ship such components, but independent verification awaits the game’s launch.
- The extent to which ML experiments might inform future gameplay systems is exploratory; claims of transformative AI‑driven reactivity are speculative until prototypes or published results are demonstrated.
Bottom line
Larian’s public stance is a cautious middle ground: adopt ML to reduce tedious work and speed iteration while publicly committing that final creative work is human‑authored and that staff reductions are not planned. The studio bolsters that claim with concrete hiring statements and the numeric detail of 23 concept artists on staff. That framing should reassure some stakeholders, but skepticism will persist among those who worry about how internal R&D choices may shift under commercial pressure.
For players and creators, the immediate takeaway is to monitor transparency: which assets were assisted by ML, how credit and compensation are handled for creative teams, and whether Larian’s exploratory uses of AI demonstrably improve design and quality without compromising authorship. The debate over generative AI in games is far from resolved, and Larian’s approach will likely be scrutinized as Divinity moves from prototyping toward production.