Today, we’re releasing a major update to Avatar Auto-Setup beta with substantial improvements to key end-to-end workflows to help you streamline your avatar creation process.
Expanded Input Model Support
Aside from the current single-mesh body input (with four unconnected components representing body+head, left eyeball, right eyeball and mouth bag), we are now also supporting:
Multi-mesh Setups: The meshes are automatically combined into a single-mesh body before being sent to Auto-Setup. The system makes a best effort to decimate the combined body parts to stay within the validation limits.
Baked Textures: Models with more than one texture map have all of their textures combined and baked into a single one. This applies to PBR textures as well, in which case four textures are baked — one for each albedo, normal, metalness and roughness.
T-Posed Bodies: Both A and T poses are now fully supported. Auto-Setup also works on I-posed bodies though we do not recommend using I-pose when the arms overlap with the torso on the rest position.
Experienced creators that prefer to rig and skin their models themselves can rely on Auto-Setup to reuse their work and just add whatever is missing to transform it into a Roblox-ready avatar.
If you provide
Auto-Setup will
R15-compatible rigged and skinned body mesh
Preserve R15-compatible rig, and generates dynamic head, cage, body partition, and attachment points
Non-R15 compatible rigged and skinned body mesh
Replace rig with a new R15-compatible one, and generates dynamic head, cage, body partition, and attachment points
Automatic Scaling
Auto-Setup will rescale your input model whenever it falls outside of the validation range. Creators will have the opportunity to undo the scaling and adjust it themselves using Studio. Auto-Setup provides visual feedback on the min and max size bounds to guide you during this process. Please note that the correct scaling is enforced on all avatars published to Marketplace.
Body parts that don’t pass size validation are highlighted in red. Use the scaling tool to adjust the size accordingly. It is possible that the proportions of your avatar fall outside the range allowed for avatars in Marketplace, in which case uniform scaling won’t fix it.
glTF Export
Auto-Setup is using glTF as its internal format to transfer information between different stages of the pipeline. Stay tuned as there are plans to enable glTF export of Auto-Setup generated avatars in the near future!
And Much More
You will also notice several UX improvements, bug fixes and an updated ML model trained on a larger variety of bodies. Please, refer to the documentation for additional information, links to templates and tutorials on how to use Auto-Setup.
We would like to invite you to experiment with this latest update on your humanoid body models. Your feedback is very important for us to continue developing Auto-Setup in a direction that meets your needs, so please share your thoughts below.
Thank you so much this is exactly what I needed! I much prefer handling weight painting myself.
In the future can we see allowing devs to provide the cage mesh themselves too? I prefer my cage meshes usually, as they sit tighter to the character and do not make clothing appear bulky on the rigs. All in all though, this is a fantastic upgrade!
These are great changes, and I’m super excited for the gltf exports, but big thing: have the symmetry issues been fixed? If so, then this is a major improvement!
To confirm, I can rig my little monster, set up a dynamic head and facial animation poses, and everything else will just happen for me? Not sure if facial setup is included in the accepted dev input.
Just wanted to chime in and say that the avatar auto-setup works really well with the bodies I’ve tested it on! Here are the things I’ve noticed while tinkering with it:
Occasionally, at seemingly random, the auto-setup will fail with an error saying that it doesn’t recognize the body or that it’s too complex. However, simply re-trying with the exact same mesh will work after a few attempts!
Facial animations are kind of a hit or miss. On some bodies they just work out of the box, on other bodies the AI seems to confuse the lower jaw, teeth & tongue with the upper jaw.
The AI really does not like extra limbs (like tails or horns). When these are kept on the main body, the caging & limb splits get all kinds of artifacts. As a result, these have to be separated into “accessories”, but UGC does not support PBR for accessories which makes it impossible to use these auto-imported avatars if they have PBR.
Cage meshes work really well. Artistically they are a bit odd on some bodies, it would be neat if we could morph / modify the cage meshes to better fit monster avatars.
Limbs are usually separated correctly and don’t usually have any noticeable artifacts, even with non-human geometry & proportions. I even tested it with a Charizard mesh (very fat nonhuman body), it worked perfectly fine!
And here's what my WIP custom avatar looks like, having been imported with the auto-setup tool!
This whole team is just… straight up cracked, the gltf export has got me so excited, right now i manually download all the cage meshes/write a script to stich it together in blender
That is genuinely so, so crazy. This is one of the best uses for AI, imo; not some shabby, ill-intentioned attempt at replacing human creativity and thought, but rather eliminating–or at least simplifying–the annoying, tedious parts that come with such creative endavours. I’m really surprised at how well the avatar auto-setup performs as well, even working on vastly different, non-bipedal rigs such as a Charizard, which someone mentioned earlier.
this is crazy 
