We are excited to announce the general release of Avatar Auto-Setup.
After the successful beta update a few weeks ago with expanded support for multi-meshes, baked textures, T-pose, auto-scaling and partially built rigs, we are graduating Auto-Setup from beta to production release.
With this general release, we would like to invite you to try and incorporate Auto-Setup in your avatar production workflows to:
Fully build avatars from scratch (i.e. from single or multi-mesh models)
Build from partially setup avatars (i.e. body or head already rigged and skinned)
What’s Next
We are committed to continue improving our Auto-Setup product throughout the new year and we plan on releasing additional features to help you achieve your creation intent, such as:
Studio Tools to edit cage meshes and attachment points
Improved ML model trained on a larger variety of bodies
glTF export of Auto-Setup generated avatars
Auto-Setup of layered accessories (including auto-caging)
Validate and fix issues (if any) on an already built avatar (i.e. auto-fix validation problems before publishing to the Marketplace)
And much more!
Your feedback is very important for us to continue developing Auto-Setup in a direction that meets your needs, so please share your thoughts and suggestions below.
Awesome to see! I’m super excited to find out what this can do.
Can you clarify, does this tool handle partially set up avatars with only a rigged head and FACS poses provided yet? My next project requires manual facial rigging because the character is not human at all, but the rest is probably human shaped enough to be auto setup.
The auto-setup feature is many, many times better than how it was at the start.
Future updates like glTF export will give us the possibilities to take the best of both worlds when making bundles, and also not have to wait for certain updates.
This is why AI is awesome, keep up the GREAT WORK!!!
I also wanted to ask, how can I support Roblox with data for machine learning like this? Is it possible for me to send over some of my own work to be trained on or is Data Sharing option the most I can do?
Would this be able to e.g. fix an avatar that’s only a handful of triangles over the limit via some kind of decimation? I had an issue with my dragon avatar where the torso was a handful of tris over the limit after using it with auto-setup despite it being under the limit in blender.
I’m working on a bundle right now that I intend to do manual rigging for, but I threw it into this tool to see how it would do. I was impressed it seemed to detect the lower jaw correctly, but there were other issues that made the result unusable.
I use sharp edges as part of my art style, but it looks like for some reason these edges are getting mangled in output, and are treated like I used the edge split modifier rather than only a trick with surface normal direction. You can see everywhere there is a sharp edge, the geometry separates. Am I doing something wrong? I’ve never had this issue anywhere else so I presume the auto-setup is doing something strange with this.
Thanks for flagging the issue (and filing a bug). We will try to get triage on it.
To set expectations, depending where the bug is, we may be unable to push a fix immediately as we are presently in code freeze until January (to keep stable over the holidays for our users). Apologies.
You have provided a great example though, and thank you for sharing the detailed pics. The head setup has a bunch of improvements planned (since the first iterations are primarily trained on humanoid heads, but is being broadened to accommodate non-humanoid over time).
Thank you Dabidar for the kind words! You have a “data share” dial when you publish an avatar or when you publish an experience. If you do that we will use it for AI training. Thank you so much!
hey PeZsmistic,
Thank you for the feedback and thank you for trying it on an amazing bird 3D model!
We are looking into the issues but please remember that Auto-Setup in its current release is only trained on humanoid characters. We are constantly expanding the range of bodies over time and this model is a great one we would want to support in the future!
Support for taking as input an already rigged head + FACS poses and setup only the body will become available early next year. We will post an update notifying when it gets out into production.
I’m sorry to hear that you hit a bug in the pipeline and ended up with a generated avatar that don’t pass validation. We want to address this and other issues to make sure all Auto-Setup generated avatars pass validation.
So yes, the auto-fix validation will resolve triangle budget for you once it is in production.
We couldn’t produce video tutorials for Auto-Setup when it was in beta since the UI and features were changing too often. But this is a good idea to do for the production release.
I will follow up with the documentation team on this.
@AndroidYouth Thanks! All good, I did not expect auto-setup to work for this character, I was just curious to see how it would do with something very unusual. For future projects though, the issue I ran into would be a problem, hence why I’m bringing it up.
Hey @HaloArbiter_101 !
I truly appreciate your thoughtful reply! I absolutely love this tool—it’s such a game-changer! Comprehensive tutorials, even as a series, would be incredible once they’re in production. I often find myself spending more time searching for answers than actually creating, and while I’m persistent, I know many might give up right before their next big breakthrough. I’m so excited about the tools you’re developing and can’t wait to see what’s next. Your effort to make things easier for creators like us is so inspiring. Thank you for everything you’re doing!
