The figure is one hierarchy now — pelvis → lumbar → chest → neck → head, with hinged elbows, knees, ankles and toes hanging off it. Every slider re-derives the whole skeleton from the head canon, so the 25 capture views can never disagree about anatomy.
Walk with WASD, Shift to run, Space to jump, Q/E to swing the camera. The spine counter-rotates against the pelvis, the head holds its gaze, and the ankles roll through toe-off.
Press 🛠️ on the camera rail, then click any imported model. Shift-click adds to the selection. Drag the gizmo, or nudge with the pads. Keys: 1 move 2 turn 3 size G group U ungroup F frame Del remove Esc clear.
The rigger looks at where a model touches the ground. It grids the contact band, finds the separate contact islands, measures each column above them, and calls each one a wheel, a leg or a post. Two legs is a biped, four is a quadruped, three or more posts is furniture, round columns are a rolling chassis. It then splits the triangles into those columns and hangs each on its own pivot, so a static GLB starts walking or rolling.
Drop a whole folder of GLBs at once. Each one gets a checkbox: tick it to put it in the world, untick to pull it out. Imports are normalised to a 2.4 m silhouette, dropped onto the terrain, and — if auto-rig is on — read for ground contacts.
Uploaded clips autoplay muted in the strip and, when ticked, become standing billboards around the hub.
Ticked tracks autoplay as a queue and feed the recorders, so whatever you hear is what lands in the file.
The inner-frame recorder grabs the render canvas with no picker. The screen recorder grabs whatever you choose to share, panels included.
Picks a clip and a track, then stretches or recuts the video so both end on the same frame. Video gets remixed, audio stays untouched.
1 Build the body as one kinematic chain, not a pile of meshes.
2 Derive every landmark from the head canon so proportions stay legal.
3 Drive the gait from a single phase: pelvis leads, spine counters, head holds gaze.
4 For imported models, read the ground contact band rather than guessing a skeleton.
5 Cluster the contacts, measure each column, classify wheel / leg / post.
6 Split the triangles into those columns and re-hang them on pivots.
7 Drive the pivots: wheels roll by arc length, legs swing in anti-phase.
8 Select, group, transform and stage the results like a modelling app.
9 Ride the CANON-25 ring around anything selected and export GLB.
10 Score the round trip on silhouette agreement, not on vibes.
Live pose drives the homunculus chain directly, so the figure is the skeleton. Assign any imported GLB as a body and it binds to that same chain — skinned for organic meshes, rigid for hard-surface. Keys: M mirror N mocap on/off.
Use an authored humanoid without re-skinning it. The original GLB skeleton, skin weights, face, clothing and animation clips stay intact. The procedural homunculus remains underneath as a fallback/control rig.
The homunculus sits inside whatever you wear. Skin weights every vertex to the nearest bone segments in the rest pose and rebinds as a SkinnedMesh. Rigid parents each part to its nearest bone — right for robots, vehicles and anything with hard joints.
bones 0 · bound verts 0 · mode —