Phase 6 updates

Phase 6 · Slice 49 ·

Skinning and clips in models

  • Mesh.boneWeights and bindposes (BoneWeight, BoneWeight1; kept as written, unsorted and unnormalized, as Unity keeps them). SkinnedMeshRenderer deforms its mesh after LateUpdate (CPU skinning into a copy it draws): each vertex and normal is the sum, by its weights as they are, of its bones' poses times their bind poses, in the renderer's space; BakeMesh with useScale (this object's space) and without (its scale kept). Scenes' and prefabs' m_Bones and m_RootBone load.
  • Model import (tassix_import): FBX skins (a bone per cluster, the cluster's mesh-to-bone bind pose converted like everything else, the four heaviest weights per vertex, normalized) and glTF skins (JOINTS_0/WEIGHTS_0 with each skinned node's joints and inverse bind matrices); FBX animation stacks baked at the file's frame rate into per-node keys. Both go through the cooked cache (version 3) and on to C#: imported prefabs' skinned renderers get their bones (the topmost as the root bone).
  • Clips inside models: a model's .meta clips (name, take, first and last frame, loop) cut from their takes, by the file IDs in its internalIDToNameTable (else Unity's hash of the name); a take name the file lacks uses its first take; one clip per take when the .meta lists none; humanoid models' clips leave the root object alone (root motion moves it in Unity). Animated models start in their first take's pose (as Unity's importer poses them). Resources.LoadAll returns a model's sub-assets by type.
  • Checked by two scenarios recorded in Unity 6000.3: skinned_meshes (weights, BakeMesh with a scaled, rotated renderer and moving bones, normals) and skinned_models (a skinned, animated FBX made in Blender for the conformance package: hierarchy, bones, root bone, weights, bind poses, rest pose, the take as a clip, poses and baked vertices as it plays; with compression off, Unity's keyframe reduction being lossy).
  • Corpus: unity-third-person's character is skinned and plays its controller's clips (idle); every clip it references resolves.
  • Not yet: GPU skinning, blend shapes in models, glTF animations, humanoid retargeting (avatars, muscles, IK); Unity's animation compression (keyframe reduction) isn't reproduced.
  • Next: Visual Scripting; URP render features; humanoid avatars and retargeting.