#xranimator

[chemb0t]

So it turns out that this system can interpret any 2d image sequence, not just live webcam feeds. I decided to test that by feeding it some animation reference clips. While it's not perfect, I'm quite impressed with how closely it tracks!

The clips are originally from Kevin Parry, which btw has an amazing library for animation reference: https://youtu.be/31VloqP-Fpo?si=LoK72RvGCZtz_OIb

[chemb0t]

Trying out some workflows to convert a VRChat avatar to Vtuber/VRM format! It's a bit convoluted but not impossible. Almost all of my blendshapes transfer over, so all of that hard work is preserved. The best thing is there's a bunch of stuff you can do with the avatar once it's a VRM, such as driving it using the latest ML research based face tracking and pose estimation tools.

What you're seeing here is eye, face, hands AND full body tracking using a single standard webcam. That's right. No iPhone, no Leap Motion, no special hardware required! This is huge because it means that anyone can make super expressive avatars without having to pay for expensive hardware to drive them anymore!