Mixamo against KayKit

Clips — bar is hand drop against an arm hanging at the side

• Mixamo
• Start-Walking3.71s · hands 0.05 from the body · a full hang+0.26
• Breathing-Idle12.46s · hands 0.05 from the body+0.26
• Running0.83s · hands 0.04 from the body · arms up, as they should be+0.15
• T-Pose0.08s · hands 0.28 from the body · the rest pose, correctly flat-0.00
• KayKit
• Walking B1.08s · hands 0.17 from the body · KayKit's best walk+0.18
• Idle1.08s · hands 0.17 from the body+0.18
• 1H Melee Attack Chop1.08s · hands 0.17 from the body · no Mixamo melee downloaded yet+0.14

Hand drop is how far below the shoulder the hand sits, in units of the character's own span, and a full bar is 0.26 — an arm hanging at the side. Mixamo's walk fills the bar. KayKit's best walk reaches 0.18, and its hands sit 0.17 from the body where Mixamo's sit 0.05. A clip whose arms are meant to be up scores low and is right to: Running is 0.15 because you run with your arms up, and T-Pose is 0.00 because that is what a T-pose is.

What you are looking at

Character

T-pose reference, 54 canonical bones, corrected bind pose

Grip

socket mode — the sword is parented to the hand bone

Mixamo

65 bones, arm span 30.8% of skeleton, bind pose flat to 0.000

KayKit

arm span 46%, and no hands at all

Ours

arm span 26.7%

Mesh here

decimated to 9k verts for the page; the real one is 120k

Why the gap closes. Nothing was tuned. KayKit is a chibi — its arms are 46% of its own skeleton's span, so an arm angle that clears a barrel chest stops at chest height on a realistic figure. Mixamo's are 30.8% against this character's 26.7%, and the pose means roughly the same thing on both bodies.

Weapon and grip come from attach.py --mode socket, motion from retarget.py, which now carries a profile per source rig — KayKit suffixes .l/.r, Mixamo prefixes Left/Right. Fingers are posed rather than keyframed, so the grip survives only because no digit is mapped: run the attach first, then the retarget.