Apply a IAutoMoviePose to a built model by running the engine's
forward kinematics and writing each bone's local rotation onto the
corresponding THREE.Bone. The pose's root transform (if any) is applied to
the model root.
The engine owns the math (semantic angles → quaternions), so the viewer only
copies results onto three.js objects, keeping rendering a thin, swappable
layer over the deterministic core.
restFrames optionally reads each joint angle as clinical and maps it
into the rig's rest-relative space before articulating (e.g. abduction 180
raises either arm overhead regardless of side); omit it to treat the pose's
angles as raw rig-space, the historical behaviour.
Morph targets that buildModel recorded as pose-driven take the weights
resolveMorphTargetWeights computes from the same pose, so a corrective
follows every applied pose; expression-written targets are left alone.
Every skinned mesh then starts a new pose revision
(invalidateAutoMovieSkinnedPose), so its bounds and vertex queries
follow the pose the shader draws.
Returns the skeleton bones the FK resolved but the model's bone map does not
carry (#1051). A deliberately partial imported map (a VRM without toes)
reports its unmapped bones every call: the host compares against what it
MEANT to map, so a typo in the bone map is distinguishable from an
intentional gap instead of both silently not articulating.
Apply a IAutoMoviePose to a built model by running the engine's forward kinematics and writing each bone's local rotation onto the corresponding
THREE.Bone. The pose's root transform (if any) is applied to the model root.The engine owns the math (semantic angles → quaternions), so the viewer only copies results onto
three.jsobjects, keeping rendering a thin, swappable layer over the deterministic core.restFramesoptionally reads each joint angle as clinical and maps it into the rig's rest-relative space before articulating (e.g. abduction 180 raises either arm overhead regardless of side); omit it to treat the pose's angles as raw rig-space, the historical behaviour.Morph targets that
buildModelrecorded as pose-driven take the weightsresolveMorphTargetWeightscomputes from the same pose, so a corrective follows every applied pose; expression-written targets are left alone. Every skinned mesh then starts a new pose revision (invalidateAutoMovieSkinnedPose), so its bounds and vertex queries follow the pose the shader draws.Returns the skeleton bones the FK resolved but the model's bone map does not carry (#1051). A deliberately partial imported map (a VRM without toes) reports its unmapped bones every call: the host compares against what it MEANT to map, so a typo in the bone map is distinguishable from an intentional gap instead of both silently not articulating.