
Model: SCAIL2 (14B)
Driving Video:
Reference Image:
Result:
Discussions: This one made a mistake of considering the front facing. I think this is because the mannequin we use does not have any facial features. If we were to use a better driving video with facial features, I think we may be safe.
Next Steps:
- Use VRM or VRChat model for driving Video and see if it resolves front-back swap issue noticed in this round.
- Compare it with One2All 14B results.
- The rendering of VROID model is very flat making it hard to debug. We will update the rendering of the reference image
- Read SCAIL 2 Paper
— Sprited Dev 🐛
Comments
Join the discussionNo comments yet. Be the first to comment.



