Thank you so much for using Auto-Setup to generate Roblox-ready avatars for your experiences and marketplace publishing. We really appreciate your support and feedback!
We are excited to announce that we just released a new round of updates to Auto-Setup consisting of the following:
Partial Input includes Dynamic Head: Users can provide a R15 rigged face and body to auto-setup, and it will reuse the rigs and produce only the missing constructs (cage, partition, attachments).
Studio Edit Tools: The generated attachment points can be edited in Studio. Users can visualize the result of their edits in the Auto-Setup preview window.
Team Create Support: The ownership of the generated assets is now set to the owner of the placefile being used instead of the user running Auto-Setup. If the placefile is owned by a group then the generated assets will belong to the group as well.
See how to edit attachments using the Studio Edit Tools in this demo.
Known issue on using blocky bodies
We have noticed that Auto-Setup is frequently failing when the input is a blocky body. This happens when the arms and legs have very low resolution and Auto-Setup is unable to find an edge-loop to partition and create elbows and knees.
In the meantime, if you can’t wait for the fix and need to Auto-Setup a blocky body, we recommend the following workaround (again, this won’t be needed after we release the fix):
Increase the resolution of the arms and legs of your blocky avatar.
Avoid having the blocky legs touching each other (add a space separating them).
The low-resolution blocky body (left) fails with Auto-Setup. The same body retopologized to a higher resolution (right) has a successful run with Auto-Setup.
What’s Next
We are committed to continue improving our AI-powered Auto-Setup product and plan on releasing additional features throughout the year to help you achieve your creation goals.
These include:
Edit of generated cage meshes
Improved support for blocky bodies
glTF export of Auto-Setup generated avatars
Auto-Setup of layered accessories (including auto-caging)
Validate and fix issues (when applicable) on an already built avatar (i.e. auto-fix validation problems before publishing to the Marketplace)
Improved ML model trained on a larger variety of bodies
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.
Team Create Support: The ownership of the generated assets is now set to the owner of the placefile being used instead of the user running Auto-Setup. If the placefile is owned by a group then the generated assets will belong to the group as well.
I primarily experience this ownership issue when uploading individual assets directly from Studios Asset Manager, especially when swapping textures on bundles. Currently, I have to go to the website to upload textures and then return to Studio, which disrupts my workflow.
Would it be possible to expand this feature to uploading assets from within studio, or better yet contain the option to switch the asset ownership behavior when uploading directly from Studio as well?
I have a blender file with a character already rigged, split into separate objects (not joined), and with facial poses. What’s the correct way for me to export this from blender so that autosetup can work on it?
I had trouble with this last time I tried last week. Not sure if I was doing something wrong or if it was the lack of the features announced today. I’ll be trying again tonight.
I love this tech, this would have been the stuff of dreams to me when I was younger.
With respect to this, I’m trying to get a custom character through auto-setup and the whole process is very frustrating to figure out.
This character is rigged for R15, both body and head with FACS poses. I’ve attempted to import it into Roblox both segmented into capped R15 body parts, and as a singular rigged mesh.
For the segmented case, I get a myriad of issues trying to get through auto-setup. For example, the below errors when imported as an R15 character, and then the tool ran on the root model.
When I try to select just the meshparts inside the model, the head is missing…
Nothing has been moderated.
And for the singular rigged mesh case, it is either imported as a custom rig and I’m not sure if studio understands where the face rig is, or it doesn’t even make it into studio with the facial animation data. Either way, the result ends up re-rigged by the tool and the FACS poses are lost.
Is there any way to insert already done caging(1) and some separated limbs(2)?
For example
(1) you weightpainted the whole body and caged it, besides the head and you want the head to be done by Roblox’s setup, BUT want the cage to stay the same
(2) You sepeated and renamed the limbs the right way but left the other half of the body unsepareted like this:
Looks like we’re going in the right direction. Now all I need is for it to accept a cage I make. Only thing I want the AI for is doing the facial animations because getting those to validate is absolute torture. There’s always one animation that’s off and it doesn’t say which one so I spend hours trying to figure it out.
Update here, we’re still investigating, but here’s some fun technical details while we look into it.
The input as a partitioned mesh lost us animations + custom properties, forcing a loss of the rig, which means we tried to re-merge and then re-partition your character.
The partitioning algorithm itself failed and labeled everything as a head, so we’re looking to get that to be more robust.
We have a working version of this asset using a different import technique if you’d like it (if it’s blocking you from something), but if you wait a bit, we’ll end up deploying a more robust pipeline in general.
This is helping us focus in on the weakest part of our pipeline, so once again, thank you for all the details!
I’m in no rush with this character, it was made for fun so I’m okay to wait. I still have to make the PBR textures anyway. I assume the different import technique is something internal you had to change.
Thanks for the details!
Getting into all of this new avatar tech has been an exercise, so I’m happy to provide problematic test cases if they can help you make the experience smoother.
Let me follow up with the Studio team on this.
Congrats team!