Move a rig.
Blend every layer.

Load the preserved rig

final rig = await GlintGlbRig.fromAsset(
  'assets/models/character.glb',
);

final animation = GlintAnimationController(
  rig,
  initialAnimation: 0,
);

GlintGlbRig keeps local meshes, nodes, skins, inverse bind matrices, and clips together. The renderer validates skin indices, joint references, finite matrices, and its per-skin joint limit before drawing.

Advance and render a pose

final frame = animation.update(dt);

return GlintGameInstance(
  model: 'character',
  animationPose: frame.pose,
  transform: characterTransform,
);

When animationPose is present it takes precedence over raw clip index/time fields. Sampling supports LINEAR, STEP, and glTF CUBICSPLINE channels.

animation.play(
  animation.animationIndexByName('Run'),
  fadeDuration: .25,
  restart: false,
);

An interrupted crossfade begins from the currently visible mixed pose, avoiding a pop back to either source clip.

Layer only the bones you need

final upperBody = GlintBoneMask.fromNodeNames(
  rig,
  ['Spine'],
  includeDescendants: true,
);

animation.addLayer(GlintAnimationLayer(
  name: 'aim',
  animationIndex: aimClip,
  mask: upperBody,
  weight: .8,
));

Override layers blend toward a clip. Additive layers apply a delta from a reference time, useful for recoil, lean, breathing, and facial motion. Timeline events report every marker crossed, including loops and reverse playback.

Extract root motion once

Configure a skeleton root and let the controller return translation and rotation deltas. Apply those deltas to the entity or physics motor; optionally remove them from the rendered pose so motion is not doubled.

final controller = GlintAnimationController(
  rig,
  initialAnimation: runClip,
  rootMotionOptions: const GlintRootMotionOptions(
    nodeIndex: 3,
    removeFromPose: true,
  ),
);

final frame = controller.update(fixedDt);
character.applyRootMotion(frame.rootMotion);

Drive animation with state

GlintAnimationStateMachine coordinates parameter-driven transitions, one-shots, and return behavior on top of the same controller. Keep gameplay authority outside the animation graph: the state machine decides presentation, not collision or game rules.

Blend animation and ragdoll physics

A GlintRagdollDefinition maps arbitrary node names or indices to colliders and fixed, revolute, or spherical constraints. It works for humanoids, animals, robots, or a partial simulated appendage.

final ragdoll = GlintRagdoll.create(
  world: physics,
  rig: rig,
  pose: animation.pose,
  simulated: false,
  definition: definition,
);

ragdoll.driveFromAnimation(animation.pose);
ragdoll.setSimulated(true, linearVelocity: character.velocity);
ragdoll.applyImpulse('chest', impactImpulse);

final renderPose = ragdoll.poseFromPhysics(
  animation.pose,
  weight: ragdollBlend,
);

Kinematic mode follows animation. Dynamic mode lets the backend solve the articulated bodies. Pose blending preserves unmapped nodes, so face, accessory, and partial-body animation can continue.