# License Game Studio Data to AI: Gameplay, Assets and Code

> How game studios, including closing studios, license gameplay with inputs, 3D assets, motion capture and source code to AI developers, and who owns what.

Canonical: https://origindatapartners.com/guides/license-game-studio-data-ai
By: Origin Data Partners
Status: Published
Published: 2026-11-13
Updated: 2026-11-13
Sources checked: 2026-11-12

## Quick answer

Yes, a game studio can license its own data to AI developers, and world-model developers want it now: gameplay video with frame-aligned controller inputs and engine state, 3D assets, animation and motion capture, and source code with history. A studio can license only what it owns. Publisher agreements, engine terms, store-bought assets, licensed music and performers' likenesses each need checking, and Unity's terms require Unity's authorization for training on material made with its software.

## Key points

- World-model developers want gameplay with the inputs recorded frame by frame, plus engine state.
- Worldmodeldata says it has licensed almost 1 million hours from studios; no price has been published.
- Store-bought assets (Unity Asset Store, Fab items tagged NoAI) cannot go into a training license without consent.
- If a publisher owns the game's IP, the studio needs the publisher's agreement.

## Can a game studio license its data to AI companies?

Yes, for what it owns. The buyers are mostly teams building world models: AI that predicts how an environment changes when someone acts in it. "For world models, you need cause and consequence," one researcher told WIRED, and games record exactly that ([WIRED](https://www.wired.com/story/the-next-evolution-of-ai-is-learning-from-your-dodgy-gaming-skills/)). Worldmodeldata "licenses footage and engine data from titles built on Unreal and Unity, rather than scraping the web." ([The Next Web](https://thenextweb.com/news/worldmodeldata-7m-seed-gaming-world-models)).

This page is for studio owners and heads of studio, including studios that are winding down. It is general information, not legal advice; your counsel decides how the contracts below apply to your games.


## What makes a gameplay dataset valuable to AI developers

Inputs aligned to frames. Worldmodeldata "packages the video, the player inputs and the underlying 3D state into clean datasets." ([The Next Web](https://thenextweb.com/news/worldmodeldata-7m-seed-gaming-world-models)). Footage alone, like a trailer or a stream, teaches far less than footage with the controller or keyboard input that caused each frame.

| Studio record | What a developer uses it for | What raises value |
| --- | --- | --- |
| Gameplay video with frame-aligned controller or keyboard inputs | World models that learn action and consequence | Many hours, many levels and situations, input logs recorded at capture |
| Engine state per frame (positions, camera, physics, events) | Ground truth that video alone cannot give | Logged from the engine, not reconstructed |
| 3D assets and scenes in source format | Generative 3D and scene models | Original source files with materials, rigs and LODs |
| Animation and motion capture | Character motion and humanoid robotics | Raw capture plus cleaned data, with performer releases |
| Source code and build history | Coding models trained on large, long-lived codebases | Full version history, tickets and reviews |
| Playtest telemetry and QA recordings | Behavior and bug-finding models | Recordings linked to the bug reports they produced |

Limits are real. Nvidia's world-model lead told WIRED: "I would be more conservative on using video game data for manipulation" ([WIRED](https://www.wired.com/story/the-next-evolution-of-ai-is-learning-from-your-dodgy-gaming-skills/)). Game data teaches movement, navigation and cause and effect better than fine hand control. Code is covered in more depth in [selling source code to AI companies](https://origindatapartners.com/guides/sell-source-code-ai-training).


## Who buys game data: suppliers, marketplaces and in-house labs

Three kinds of company are active, and none publishes prices.

| Company | What it does | What is public |
| --- | --- | --- |
| Worldmodeldata (UK) | Licenses gameplay and engine data from studios and resells it to world-model labs | Says it has licensed almost 1 million hours from studios it does not name ([WIRED](https://www.wired.com/story/the-next-evolution-of-ai-is-learning-from-your-dodgy-gaming-skills/)); in July 2026 it had "no finalised customer contracts and no revenue" ([The Next Web](https://thenextweb.com/news/worldmodeldata-7m-seed-gaming-world-models)) |
| Origin Lab (US) | A marketplace between game companies and world-model labs | Raised "an $8 million seed funding round led by Lightspeed Ventures" ([TechCrunch](https://techcrunch.com/2026/05/13/origin-lab-raises-8m-to-help-video-game-companies-sell-data-to-world-model-builders/)). Not affiliated with Origin Data Partners. |
| General Intuition (US) | Trains its own models on gameplay clips from its sister company Medal | Raised $220 million at a $6.2 billion valuation ([GamesBeat](https://gamesbeat.com/general-intuition-raises-another-220m-at-6-2b-valuation-for-training-models-with-game-data/)); "keeps its data in-house" ([The Next Web](https://thenextweb.com/news/worldmodeldata-7m-seed-gaming-world-models)) |

Origin Lab, the game-data marketplace, is a separate company with a similar name. Origin Data Partners is not affiliated with it.

General Intuition's valuation is an investment round, not a price paid for anyone's data. It shows how much investors expect action-labeled gameplay to be worth, nothing more.


## Who owns game assets: publisher, engine, stores, contractors and players

Start with the publishing agreement. If a publisher funded the game and took the IP, the studio cannot license the game's code, assets or footage without the publisher. Then work through the layers below.

| Layer | What the terms say | What to do |
| --- | --- | --- |
| Your own game on Unreal Engine | "you own all rights, other than rights in the Licensed Technology, in the Products you develop" ([Unreal EULA](https://www.unrealengine.com/eula/unreal)) | License your game content; leave Epic's engine code and content out, and read the EULA version you accepted for any AI terms |
| Your own game on Unity | "Your User Content is yours." But Unity restricts use of its offerings or derived data "for the training, validation, or development of machine learning, artificial intelligence, or similar algorithmic models" without prior authorization ([Unity Terms of Service](https://unity.com/legal/terms-of-service)) | Ask Unity in writing before licensing anything made with Unity beyond your own source code, and keep its answer |
| Unity Asset Store purchases | No use "for purposes such as training an artificial intelligence or machine learning model without the express consent from the Provider and/or Unity." ([Unity Asset Store EULA](https://unity.com/legal/as-terms)) | Remove bought assets from what you license |
| Fab purchases | "The NoAI meta tag indicates that an asset must not be used for generative AI data collection." ([Epic Fab documentation](https://dev.epicgames.com/documentation/en-us/fab/licenses-and-pricing-in-fab)) | Remove NoAI items; check each listing's terms for the rest |
| Megascans | Epic is updating terms "to allow developers to train generative AI models on Megascans and Megaplants content." ([CG Channel](https://www.cgchannel.com/2026/11/epic-has-made-megascans-free-again-for-good-this-time/)) | Allowed, but anyone can download them free, so they add little value |
| Middleware and licensed music | Governed by each vendor's and label's license, usually limited to the game | Strip licensed audio and middleware code |
| Contractor art and code | Employees' work is a work made for hire; a contractor's is not, unless a written agreement assigns it ([17 U.S.C. 101](https://www.law.cornell.edu/uscode/text/17/101)) | Find the assignment clause in each contractor agreement |


## Motion capture, voice actors and recorded players

People in your data have their own rights. SAG-AFTRA's 2025 Interactive Media Agreement includes "consent and disclosure requirements for A.I. digital replica use" ([SAG-AFTRA](https://www.sagaftra.org/contracts-industry-resources/interactive/2025-interactive-media-video-game-agreement)). A motion capture dataset or voice recordings from union performers should not go into a training license without checking the performer's contract and getting new consent where it is required.

Player recordings are personal data. Playtest videos with faces and voices, voice chat and account telemetry need consent or removal. If players under 13 may appear, the COPPA Rule counts "A photograph, video, or audio file where such file contains a child's image or voice" as personal information ([16 CFR 312.2](https://www.law.cornell.edu/cfr/text/16/312.2)). The safe default is to license gameplay captured by your own staff and testers who signed releases.


## Sell game assets and 3D models on marketplaces, or license the whole archive?

They are different businesses. Selling 3D models or asset packs on a store such as Fab or the Unity Asset Store earns per-sale income from other developers, under the store's standard license. Licensing an archive to an AI developer is one negotiated contract for training rights to a large body of work: years of gameplay, source files and code.

| Question | Asset marketplace | AI training license |
| --- | --- | --- |
| Who pays | Individual developers buying packs | One AI developer or data supplier |
| What is sold | Finished, game-ready assets | Raw and source data with history and inputs |
| Training rights | Stores let creators restrict AI use (for example Fab's NoAI tag) | The whole point of the contract |
| Price | Per pack, set by you | Negotiated; no public price list |

A studio can do both, but an exclusive AI license may restrict later store sales of the same assets. See [exclusive vs non-exclusive data licenses](https://origindatapartners.com/guides/exclusive-vs-non-exclusive-data-license).


## What happens to game assets when a studio closes?

They stay owned by whoever held the rights: the studio's company (and so its creditors, in a bankruptcy or liquidation) or the publisher. What usually gets lost is the data itself, when cloud accounts, build servers and capture rigs are shut down before anyone copies them.

- Before switching anything off, export the repositories with full history, the asset source files, capture sessions and input logs, and playtest recordings.
- Collect the contracts that prove ownership: publishing agreements, contractor assignments, performer releases and store licenses.
- Do not delete player data needed for legal holds; do delete or exclude it from what you license.
- Assets and data can be sold or licensed as part of the wind-down; see [selling company data after closing](https://origindatapartners.com/guides/sell-company-data-after-closing) and [exporting records before closing](https://origindatapartners.com/guides/export-company-records-before-closing).


## How to prepare a game studio archive for licensing

Prepare a description first; no files need to leave the studio for a first conversation.

- List titles, engine and version, years, platforms, and who owns the IP of each title.
- Estimate hours of gameplay you can capture or already hold with input logs, and which engine state you can log.
- Inventory third-party content per title: store assets, middleware, licensed music, performer captures.
- If you build on Unity, write to Unity for authorization before anything made with Unity is licensed.


## What AI companies pay for game data: the honest evidence

No price for licensed gameplay hours has been published, and no closed sale between a named studio and a buyer has been disclosed with its value.

| Evidence | What it says | Source grade |
| --- | --- | --- |
| Worldmodeldata | Almost 1 million hours licensed from unnamed studios; no price | Named press (WIRED) reporting the company's own statement |
| Mercor game studio listing | [AIData.Broker](https://aidata.broker/guides/sell-game-data-to-ai-companies) reports the listing at "$100,000 to $1 million per company" for records, code and footage together | Third-party guide describing a buyer's published range; Origin could not open the listing itself; a range, not a deal |
| General Intuition | $6.2 billion valuation, trained on Medal gameplay clips | Named press; an investment, not a data price |

For motion capture and video rates more broadly, see the [video data prices tracker](https://origindatapartners.com/guides/video-data-prices-tracker) and [video data buyers compared](https://origindatapartners.com/guides/video-data-buyers-compared).

Does your company hold records like these? [See what your data is worth](https://origindatapartners.com/check-fit).

Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes ([how Origin is paid](https://origindatapartners.com/referral-disclosure)). [See what your data is worth](https://origindatapartners.com/check-fit).


## Questions and answers

### Can our game studio license gameplay footage to AI companies for world model training?

Yes, if the studio owns the game and the footage. World-model developers want gameplay recorded with frame-aligned controller or keyboard inputs and engine state. Leave out licensed music, store-bought assets, union performers without fresh consent, and players who did not agree. If a publisher owns the IP, it must agree first.

### What happens to game assets when a studio closes?

Ownership stays with whoever held the rights: the studio company, its creditors in a bankruptcy, or the publisher. The data is what gets lost, when servers and cloud accounts are shut down. Export repositories, source assets, capture sessions and contracts before anything is switched off, so the archive can still be licensed or sold.

### Can a studio include assets bought on the Unity Asset Store in an AI training license?

No, not without consent. The Unity Asset Store EULA bars using assets for training an artificial intelligence or machine learning model without the express consent of the provider and/or Unity. Remove bought assets, and on Fab remove anything tagged NoAI, before licensing a project.

### Does Unity need to approve licensing data from a game built in Unity for AI training?

Its terms say prior authorization is needed. Unity's Terms of Service say your user content is yours, but restrict using Unity offerings, or data derived from them, for training machine learning or AI models without prior authorization. Ask Unity in writing before licensing anything made with Unity beyond your own source code.

### Where can I sell 3D models, and is that the same as licensing them to AI?

Asset stores such as Fab and the Unity Asset Store sell game-ready packs to other developers, per sale, under a standard license. Licensing to an AI developer is a separate negotiated contract for training rights to a large archive. Stores let creators restrict AI use, so the two can be kept apart.

### Is Origin Lab the same company as Origin Data Partners?

No. Origin Lab is a San Francisco marketplace that helps game companies license data to world-model builders; TechCrunch reported its $8 million seed round in May 2026. Origin Data Partners is a separate, independent data partner and is not affiliated with Origin Lab.

### How much do AI companies pay a game studio for its data?

No price for gameplay hours has been published. A third-party guide reports a Mercor listing range of $100,000 to $1 million per studio for records, code and footage together, which Origin could not confirm on Mercor's own page. Worldmodeldata says it has licensed almost 1 million hours without stating what it paid.

### Can a motion capture dataset from our game be licensed for robotics AI training?

Possibly, if the studio owns the capture and the performers agreed. SAG-AFTRA's 2025 Interactive Media Agreement includes consent and disclosure requirements for AI digital replica use, so check each performer's contract and release before including their captures. Raw and cleaned capture with full metadata is worth more than finished animation.

## Sources

- [WIRED: The next evolution of AI is learning from your gaming skills (28 Sep 2026)](https://www.wired.com/story/the-next-evolution-of-ai-is-learning-from-your-dodgy-gaming-skills/)
- [The Next Web: Worldmodeldata raises £7M (6 Jul 2026)](https://thenextweb.com/news/worldmodeldata-7m-seed-gaming-world-models)
- [TechCrunch: Origin Lab raises $8M (13 May 2026)](https://techcrunch.com/2026/05/13/origin-lab-raises-8m-to-help-video-game-companies-sell-data-to-world-model-builders/)
- [GamesBeat: General Intuition raises $220M at $6.2B valuation (29 Sep 2026)](https://gamesbeat.com/general-intuition-raises-another-220m-at-6-2b-valuation-for-training-models-with-game-data/)
- [Unreal Engine End User License Agreement](https://www.unrealengine.com/eula/unreal)
- [Unity Terms of Service](https://unity.com/legal/terms-of-service)
- [Unity Asset Store Terms of Service and EULA](https://unity.com/legal/as-terms)
- [Epic Developer Community: Fab licenses and pricing (NoAI tag)](https://dev.epicgames.com/documentation/en-us/fab/licenses-and-pricing-in-fab)
- [CG Channel: Epic has made Megascans free again (6 Nov 2026)](https://www.cgchannel.com/2026/11/epic-has-made-megascans-free-again-for-good-this-time/)
- [SAG-AFTRA: 2025 Interactive Media (video game) Agreement](https://www.sagaftra.org/contracts-industry-resources/interactive/2025-interactive-media-video-game-agreement)
- [16 CFR 312.2: COPPA Rule definitions](https://www.law.cornell.edu/cfr/text/16/312.2)
- [17 U.S.C. 101: definitions (architectural work, work made for hire)](https://www.law.cornell.edu/uscode/text/17/101)
- [AIData.Broker: selling game code and game data to AI companies (third-party guide)](https://aidata.broker/guides/sell-game-data-to-ai-companies)

## Related reading

- [Can a company sell its source code to AI companies for training?](https://origindatapartners.com/guides/sell-source-code-ai-training)
- [Where to sell video footage for AI training: Troveo, Protege, Defined.ai, Shutterstock and Wirestock compared](https://origindatapartners.com/guides/video-data-buyers-compared)
- [AI video training data prices and deals tracker: per minute, per hour and per deal (2024 to 2026)](https://origindatapartners.com/guides/video-data-prices-tracker)
- [Closing a software business: what could have value beyond the code?](https://origindatapartners.com/guides/sell-company-data-after-closing)
- [Cancel Google Workspace or delete Slack? Export data first](https://origindatapartners.com/guides/export-company-records-before-closing)
- [Motion capture jobs: film and game performers vs paid motion data for robots and AI](https://origindatapartners.com/physical-ai/motion-capture-jobs)
- [Exclusive vs non-exclusive data licenses: what sellers should know](https://origindatapartners.com/guides/exclusive-vs-non-exclusive-data-license)