Hi Creators,
Our goal is to enable the creation of anything, anywhere, by anyone. We’ve been building generative AI tools such as Assistant and Avatar Auto-Setup to help reduce the barrier to creation for new creators and empower all creators to be more productive. We believe that AI will ultimately play an invaluable role in supporting your work. This is why we want to continue improving our technology to build smarter, more contextual AI models that understand everything about creating on Roblox – from coding in Luau to the structure of the DataModels to Avatar systems and more.
As part of this effort, we’re launching an AI data-sharing program that allows you to set your preferences over how your anonymized creation data is used to train our generative AI models. Our goal is that AI continues to accelerate creation and complement creators’ skills. Our AI tools can’t decide what the best experience or avatar is, or what will resonate with users. Ultimately, that is up to the creator – to apply their skills, intuition, and data using tools like Analytics to help them design the best content.
How the program will work
-
All experiences and avatar items (including bodies, heads, clothing, and accessories) created prior to July 10, 2024, will be set to generative AI sharing OFF.
-
Starting July 10, 2024, when you save or publish a new experience or avatar item, you will see a new toggle that allows you to control what you share to improve generative AI. The toggle is set by default to ON to allow data sharing; you can choose your preference for this asset here, and can change the default for all new experiences and avatars in Creator Hub.
-
Once you set your default sharing preference in Creator Hub, future publishes for new content of that asset type will default to this preference, but you can choose the appropriate sharing preference when you save or publish the asset.
-
When creating or updating experiences via the Publish as/Save as, if the originating experience is set to sharing OFF and/or the target experience is set to sharing OFF, it will remain OFF. Data sharing will be set to ON only if both the originating and target experiences have been manually set to ON. However, you can change this setting at any time.
-
We won’t collect or use any of your shared data to train generative AI models until September 10, 2024.
-
All creators will be able to use our generative AI tools, regardless of your data sharing preference.
-
We will continue to use creator data for the purposes of improving key platform systems such as safety, moderation, translation, and recommendations.
Revised Supplemental AI Terms
We’re also rolling out an updated AI Supplemental Terms which you will be asked to accept before Studio launches. These terms outline the rules governing the use of our AI tools. Accepting the Supplemental Terms does not affect your ability to change your sharing preferences.
At a high level, the terms address use restrictions such as not using AI tools in ways that could harm others or break the law. Additionally, as explained in the terms, the tools may not work perfectly, so use them responsibly and at your own risk.
Why we are doing this
Over the past year, we’ve rolled out a suite of generative AI-powered solutions, including Code Assist, Material Generator, Assistant, Texture Generator, and Avatar Auto-Setup. We’re also working toward 4D generative AI, going beyond static 3D objects to dynamic creations that can move and interact with their environments. More and more creators are using these solutions to learn and create faster, eliminating mundane and repetitive tasks and complementing their existing skill sets.
Some of the feedback we’ve heard from you on how these solutions are helping you be more productive are:
-
Code Assist: Toya Play saw a 40% increase in efficiency using Code Assist, representing “months of work across the company.”
-
Avatar Auto-Setup: Creators told us that previously rigging, caging, skinning, and segmenting their avatars could take days, even weeks, and now it happens in a couple of minutes. Creators also were previously intimidated to build avatars due to the complexity and have now expressed interest in building.
Your contributions to generative AI training help us create and fine-tune our AI models in a way that is sustainable and scalable for our community of creators, and will help improve creation in several ways, including:
-
More contextual Code Assist and Assistant: Experience data will help improve our AI solutions, so they can provide better and more contextually-aware suggestions for your work.
-
More flexible Avatar Auto-Setup: A more diverse avatar dataset will help it handle a wider range of avatar shapes saving you days of time on tedious setup.
Set your default preferences
Starting today, you will be able to set your preferences. In Creator Hub, you will click on your user name and select ‘Settings’ and then ‘Data Sharing’ to set your preferences for experiences, avatar items, and paid Creator Store 3D models, which we aim to launch later this year. We’ll post separately when you can sell 3D models in Creator Store, but your default preference can be set in advance. Check or uncheck the “Make sharing data my default option when publishing” option to choose your default sharing settings for each asset type.
Any creations you publish will respect your preference, and regardless of the default setting, you can always change your mind at publish time.
Free Creator Store meshes, images, and models will be included in generative AI training, without the ability to remove from training. We will not be training on plugins, audio, or videos.
We won’t collect or use any of your shared data to train generative AI models until September 10, 2024.
Express your preference
When you publish a new experience, upload an avatar item, or upload a paid model (coming later) to the Creator Store, you will have the option to share your data with our AI models. The default will be based on the preference you set in the Creator Dashboard settings. In the example below for publishing a new experience, you can see the toggle for ‘Data Sharing.’ Similar workflows exist for avatar items and will exist for paid Creator Store models once live.
Group owners control sharing
Only group owners can opt-in to share data with generative AI models. Also, on the Creator Hub Data Sharing page, group-owned content only appears in the group owner’s dashboard.
Updates to the previous Luau Script Data Sharing program
Last year, we launched a data sharing program that allowed you to share your scripts with Roblox to improve our AI models or with an open-source-available Luau Dataset designed to help all LLMs and AI models understand Luau. Going forward, we will not use the script data previously shared with Roblox only. We appreciate everyone who shared their scripts as it helped us improve the effectiveness of Code Assist. Instead, you will have the opportunity to decide whether you want to share your experience data, including your script data, with our new data sharing program.
Sharing your data for the Luau public dataset is not affected by this program. You can set your preferences in Creator Dashboard under the Public Luau tab. However, opting out from the Luau Dataset won’t be possible after 30 days as the dataset will be publicly available. Please only contribute code from experiences that you are comfortable with. Only a small fraction of your contributed code will be compiled and aggregated into an open source-available dataset for AI model training.
Building Safe and Responsible Generative AI Models
We take the security of your data seriously. All collected data is processed, stored, and shared with the utmost care, adhering to stringent security protocols. All data for AI training undergoes a rigorous cleaning process to remove sensitive and/or potentially harmful information. Additionally, the data is split up into tokens (small snippets) for training, similar to paper shredding. We shared a post on how we’ve approached this for script training and plan to share more for other asset types going forward.
Thank you for being a vital part of the Roblox community!
Overview of Data Sharing by Asset Types
Asset Type | Data Sharing Preference |
---|---|
Experiences and Avatar Items created prior to July 10, 2024 | Sharing OFF by default, control in Creator Dashboard Settings, future updates of this content will remain as OFF unless you decide to update your preferences |
Experiences, Avatar Items, Paid Creator Store Models* created since July 10, 2024 *Coming later this year |
Sharing ON by default, control in Creator Dashboard Settings |
Free Creator Store Assets (Meshes, Images, Models) | Included in generative AI training, no ability to set data preference |
FAQs
Click here to view the FAQ!
Why should I participate in the data-sharing program?
- Your participation in the data sharing program will improve the quality of AI creation tools, which will make it faster and easier to bring your ideas to life.
Will I still be able to use generative AI tools if I don’t share my data?
- You can still use all of our AI products even if you don’t share your data.
Does accepting the Supplemental Terms in Studio mean that I’m sharing my creation data for generative AI training?
- No. Accepting the Supplemental Terms does not affect your ability to change your sharing preferences of your creation data. This is controlled by your preferences in Creator Dashboard and when you publish or upload an asset.
When will you start using the new data shared?
- Please note that the new policy will become effective on September 10, 2024. Until then, Roblox will not begin collecting any creation data for AI training, except scripts that have already been shared through our existing data sharing program for Luau scripts.
Will you use my existing avatar items or experiences for AI training?
Any experience or avatar item that you save/publish before 7/10/2024 will not be used for AI training * unless you explicitly update your data-sharing for a specific creation in Creator Hub or when publishing or uploading an update to that creation.
How are my data preferences affected if I use the Publish As or Save As features?
-
When creating or updating experiences via the Publish as/Save as, if the originating experience is set to sharing OFF and/or the target experience is set to sharing OFF, it will remain OFF. Data sharing will be set to ON only if both the originating and target experiences have been manually set to ON. However, you can change this setting at any time.
Note: Currently, when you publish an existing place file into an existing experience, you will receive a warning that your preferences may change, as we will follow the preference expressed in the target experience. We are working on updating this to only set to sharing ON if both the source experience and target experience are set to sharing ON.
Why is Roblox developing its own generative AI models / Why not use off-the-shelf models?
- Roblox is a unique platform in that creators are building immersive, multiplayer 3D worlds. To create solutions that solve your unique needs, we need AI models that understand the context of Roblox’s platform, from interactive avatars to the entire data model. We also saw from our previous data-sharing program for scripts that fine-tuning with community-contributed data helped improve the effectiveness of our code suggestions in Code Assist. Last, using our own AI models is more cost-effective than using off-the-shelf third-party models and thus allows us to provide wider access to a larger number of creators.
I don’t see the ability to control my AI Data Sharing
- You can set your data sharing preferences in Creator Hub. The interface in Studio for managing sharing will be rolling out this week. We won’t collect or train on any of the data that you choose to share until, at the earliest, September 10, 2024. Creators under 13 will not have the option to share their data and will not see the UI for data sharing.
At RDC 2023, you said that you wouldn’t use your script data without our permission
- At publish time, you will be given the option to set your data sharing preferences. We will train on this data only after you submit this preference to allow sharing. Additionally, you have the ability to set your default preferences in Creator Hub.
Will you sell my data to 3rd parties?
- We will never sell your data to a third party. We are only interested in helping you become more successful on Roblox more quickly.
Why can’t I set a data-sharing preference for free Creator Store items such as meshes, images, and models?
- Since these assets have already been shared and are freely and publicly available for creators, we’ve decided to include them in the AI data training. You can remove free models, images and meshes from the Store before September 10th, 2024, if you don’t want us to use them in training. However, if these free assets were already added to experiences, they may be subject to training. We will not be training on plugins, audio or video. We won’t use any of your non-public assets unless you’ve chosen to share an experience that includes those assets.
Does data sharing apply to both free and paid Avatar items in Marketplace?
- For generative AI training, both free and paid avatar items are treated the same.
What happens if I change my preference and decide to stop sharing one of my creations?
- You can stop sharing at any time. Your data will be removed from the raw data set in less than 30 days, and the change will be applied in our next AI model, which we will refresh at least once a year.
What do I do if I see something that appears to be a copy of my work?
- As always, if you see content on Roblox that you believe infringes on your copyrights, you can always report that content to us for removal under the DMCA. Rights Manager allows you to file a removal request when you find content that infringes on your creations and intellectual property. If you have more questions about IP and generative AI, we’ve put together a primer on the most common questions we’ve heard about protecting your intellectual property when creating with generative AI.
The Supplemental AI Terms mention including third-party services. Can third-party services train on my data?
- Except for the public Luau dataset you agreed to share publicly, we will not permit third parties to train on your data.
The Supplemental AI Terms mention machine-learning training on my inputs, what does this mean?
- We use your anonymized interactions with Studio including your inputs to and outputs from our generative AI tools for training, to improve the effectiveness of the tools we provide you. For instance, your thumbs-up and thumbs-down rates with corresponding outputs, and command bar prompts help us better fine-tune our solutions.