[Update] September 25, 2025
Hi Creators!
We are excited to announce new APIs for an early preview of our Photo-to-Avatar technology. We’re building new generative AI features that allow developers to create unique experiences and this is an early look at what we’ve been working on!
These APIs allow you to turn a user-provided selfie into a fully rigged, FACS-enabled, skinned Roblox avatar that users can equip in your experience, and when combined with existing APIs, edit, pay for, and save to the platform. They are designed to help you build a proof of concept for testing photo-to-avatar generation capabilities, and determine how best to integrate it in your experiences.
These APIs are available now, and over time, we will continuously improve them to more accurately represent your photos and provide additional ways to customize and style your avatars.
How it Works
The Photo-to-Avatar API allows you to initiate a flow where users can create an avatar from a selfie. When paired with PromptCreateAvatarAsync, users may pay for and save the avatar to their inventory to be used across the platform. As the experience creator, you receive your share from the in-experience sale and avatars created in experiences have attribution back to that experience.
Creating an avatar from a photo and text prompt is a multi-step process. Here’s an overview of the workflow (see full documentation for Photo-to-Avatar generation and AvatarCreationService APIs)
-
Request an AvatarGeneration Session - First, you’ll need to request an
AvatarGenerationsession for the player on the server usingAvatarCreationService:RequestAvatarGenerationSessionAsync(). This provides the player with a set number of avatar preview, avatar generations and time limit for the generation session. -
Prompt the User for a Selfie & Generate a 2D Preview - Next, you’ll prompt the user to take a selfie using
AvatarCreationService:PromptSelectAvatarGenerationImageAsync()to get afileIdstring. You’ll then use thisfileIdalong with theSessionIdand an optional text prompt to generate a 2D preview of the avatar by callingAvatarCreationService:GenerateAvatar2DPreviewAsync()on the server. The returnedpreviewIdcan be sent to the client to retrieve the preview as anEditableImageusingAvatarCreationService:LoadAvatar2DPreviewAsync().
- Generate the Full 3D Avatar - Once the user is happy with the 2D preview, you can generate a full 3D avatar by calling
AvatarCreationService:GenerateAvatarAsync()on the server. This method takes a dictionary containing theSessionIdandPreviewIdas input and returns agenerationId. You can then retrieve the generated avatar as aHumanoidDescriptionusing theAvatarCreationService:LoadGeneratedAvatarAsync()method. The mesh and textures are provided asEditableMeshandEditableImageobjects, allowing for further editing.
- Edit and Save the Avatar to Inventory - After the user has had a chance to edit the avatar, they can publish it using the
AvatarCreationService:PromptCreateAvatarAsync()method. Learn more about saving an avatar to inventory
To provide a clear understanding of how the code is implemented, we are linking the place file. This will allow you to examine the implementation firsthand and use it as a starting point to build and test your own experience.
Current Limitations
We know that the outputs are limited at this time, but wanted to provide this as an early preview to get your feedback and allow you to start experimenting with the APIs. We will continue to improve these APIs, and this alpha release has some current limitations:
- Platform availability: This feature is currently only available on iOS and Android.
- User eligibility: This release is limited to users who are 13+ in select countries.
- Selfie-only: Users can only take a live selfie; uploading from the camera roll is not supported at this time.
- Output quality may vary: The avatars that this initial version generates may not always resemble the original image – lighting, backgrounds, and other variables may have unexpected effects on the generated avatar. We know that the quality of avatars this early model can generate is limited, and we’re actively working on iterating and improving the output quality to better match the input image.
- Realistic stylization only: Currently, the generated avatars are only of a more realistic aesthetic.
- Limited to head and body generation: The API generates a head (with optional hair) and a body. The body type generated may be different from the person’s actual body type as it’s only generated based on the photo of the head. It does not generate clothing or accessories.
We plan to extend the API in the future to enable a wider range of aesthetics, availability on more platforms, a larger range of possible outputs, and the generation of clothing and accessories. This release is a crucial first step toward that goal, and we look forward to seeing what you create and hearing your feedback!
Special thanks to the team that helped make this happen: @hol0gr4ph, @squickeman, @haomiaojiang, @chickenbasil42, @ubluec,@generalcourse, @TheGamer101, @RadioNoggin, @whol8wh8at, @CackleAtGrackle, @moonstaryoon, @FarazTheGreat, & @timetopretend8.
Frequently Asked Questions [updated 09/17/25]
How are developers paid?
- When a user makes a purchase, developers will receive 70% (30% as ‘creator’ and 40% as the experience developer) of sales that come from in-experience creation.
Can users sell their avatar creation?
- Currently, we are not allowing generated avatar content to be sold on the Marketplace. In-experience created bodies go to the user’s inventory, and cannot be sold on Marketplace, traded, or resold to other users.
What are the developer requirements?
- The requirements for developers here are the same as they are for UGC avatar creators: you must be ID-verified and Premium to allow the implementation of the publishing APIs. Additionally, developers will have to buy a token tied to a UniverseID and asset type with a price point. Developers pay for this token and a publishing fee, similar to the process for avatar creators.
Do users need to be ID-verified to publish?
- Users do not need to be ID-verified or have Premium to create an avatar in an experience.
If an avatar fails moderation, does the user receive a refund?
- No, if the avatar violates Roblox community standards, the user will not be able to receive a refund. As with other submissions, they may appeal the decision if they believe their avatar was incorrectly moderated and should be approved.
How did you train these data sets?
- We trained our model on datasets that we compiled and developed and did not train on any data from Roblox creators.
How did you address potential bias in the datasets?
- We recognize that AI algorithms and datasets can potentially reflect or amplify unfair biases. Addressing and removing these biases from our models is a key priority for Roblox. To guard against bias in our own models, we leveraged diverse data sets, insights, and feedback to finetune our AI. We also extensively tested for violative content, including developing new tooling to proactively catch and address potential violations.
Will uploading my selfie for this feature get me banned or warned?
- No, you will not violate any terms by uploading a selfie when using this product. Our Community Standards have the following statement. “To maintain privacy and safety, users may be prohibited from sharing or requesting personal information on Roblox depending on their age, including: Visual and audio media of themselves or other non-public figures”. Among other things, our community standards prohibit unauthorized sharing of personal information between users. In the case of Photo-to-Avatar, this information is being shared with Roblox to provide the feature and we’re specifically requiring consent. In addition, we are restricting this feature based on age (13+).
