agentleFS
EL4CTEOEL4CTEO/blender-skills/skills/blender-animation/SKILL.md

blender-animation

Skill779 starsChanged 15 days agoMITView on GitHub

Animation in Blender 5.2 - slotted/layered actions and the channelbag API, Python keyframing, F-curves, game loops with pop/hitch checks, NLA clips, pose library, mocap/BVH/Mixamo retargeting with auto bone maps, baking control rigs to export skeletons, root motion, per-clip export. Use for any keyframe, action, NLA, mocap, or animation-export task.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Breakdowns: add in-betweens that define arcs and spacing.

    line 22
  2. Distinctive:

    Keep the bone transform channels the engine expects.

    line 33
  3. Very specific:

    Import: bpy.ops.import_anim.bvh(filepath=..., global_scale=0.01 if cm else 1, rotate_mode="NATIVE", update_scene_fps=True), …

    line 43

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Animation

The execution model is in blender-core (its references/bpy-patterns.md has verified keyframe/driver snippets). Rigs are in blender-character-rigging.

5.x data model (what changed)

  • An Action holds slots (one per animated ID) and layers → strips → channelbags → F-curves. A single action can animate several objects (e.g. a rig and its camera) through different slots.
  • Assign: ob.animation_data.action = act, then make sure ob.animation_data.action_slot is set (picked automatically when a suitable slot exists; otherwise act.slots.new(id_type="OBJECT", name=ob.name)).
  • F-curves: cb = anim_utils.action_ensure_channelbag_for_slot(act, slot), then cb.fcurves.ensure(path, index=i, group_name=...). action.fcurves no longer exists.
  • The simplest route is still pb.keyframe_insert("rotation_quaternion", frame=f), which creates the action, slot and channelbag.

Workflow (keyframe animation, studio style)

  1. Blocking: key poses on every channel (CONSTANT interpolation, bpy.context.preferences.edit.keyframe_new_interpolation_type) on golden frames (contact, down, passing, up for walks).
  2. Breakdowns: add in-betweens that define arcs and spacing.
  3. Spline: switch to BEZIER, fix spacing in the Graph Editor, and clean overshoot. Auto-clamped handles avoid overshoot.
  4. Polish: offsets/overlap (follow-through on hair, tails, cloth bones), arcs, moving holds (never a perfectly frozen character), and contact stability (feet locked with IK).
  5. Review: MCP look(mode="frames", frame_count=8) or a Workbench playblast (blender-rendering).

Game animation rules

  • One action per clip, named <Rig>_<Clip> or just <Clip>, with a fake user so it survives saving.
  • Loops: first and last frames identical. Export frames 1..N and let the engine drop the duplicate, or exclude the last frame (UE/Unity "loop pose" options). An F-curve CYCLES modifier helps preview but doesn't export, so bake it. Verify with scripts/loop_check.py -- --object RIG --action Walk [--root-motion root] [--fix]: it reports pose pops (last ≠ first, per bone, in deg/m) and speed hitches across the seam, and --fix keys the first-frame values on the last frame (root travel is kept).
  • FPS: author at 30 (or 24) and let engines resample. Set scene.render.fps before importing mocap.
  • Root motion: the root bone carries the translation (and yaw). The hips keep the bob/sway. scripts/bake_action.py --root-motion xy moves the hips' ground motion to the root. In-place clips: the root stays at the origin.
  • Additive/overlay clips (aim offsets, breathing) are exported as normal clips and marked additive in the engine.
  • Keep the bone transform channels the engine expects. Don't animate bone scale unless the engine supports it (Roblox and many mobile pipelines don't).

Bake control rig → export skeleton

scripts/bake_action.py -- --list
scripts/bake_action.py -- --source RIG_Body --target GAME_RIG_Body [--actions Walk Run] [--root-motion xy --hips spine]

This gives one baked action per clip (GAME_RIG_Body|Walk), cleaned, with fake users, plus one muted NLA track per clip (CLIP_Walk) so FBX (bake_anim_use_nla_strips/all actions) and glTF (ACTIONS mode) export every clip. The target's constraints are muted afterward so it plays the baked data; run --follow to drive it live again. Self-bake (IK/constraints → keys on the same rig): --source X --target X.

Mocap and retargeting

  • Import: bpy.ops.import_anim.bvh(filepath=..., global_scale=0.01 if cm else 1, rotate_mode="NATIVE", update_scene_fps=True), import_scene.fbx (Mixamo, automatic_bone_orientation=True).
  • Retarget: scripts/retarget.py -- --source Mocap --target GAME_RIG [--map auto|map.json] [--in-place] [--name Walk]. Each target bone gets the source bone's world rotation change from rest on its own rest orientation, so rolls, bone axes, Y-up/cm FBX object transforms and proportions don't matter. The hips travel is scaled by the hips-height ratio. Keys go straight into a new TARGET|Name action (quaternion, linear, fake user). Measured on a Mixamo-style rig (Y-up, 0.01 scale, rolled) → Rigify-named rig with 15% longer legs: 17 bones, < 0.1° error.
  • The auto map understands Mixamo (mixamorig:LeftUpLeg), Rigify DEF-/ORG-, UE/Unity/BVH-style names, and spine chains of different lengths. Check it first with --show-map (unmapped_key_parts must be empty), and pass a JSON {target: source} for anything custom (fingers, props).
  • Rest poses must match (T vs A matters most): pose the target's rest like the source's first. Then clean, add root motion, and export with bake_action.py.
  • Clean mocap: decimate keys (graph.decimate, ratio 0.3–0.5), use an F-curve SMOOTH modifier or the Butterworth filter (graph.butterworth_smooth), and fix foot sliding with IK pins and then bake.
  • Production retarget tools: Auto-Rig Pro (paid), Rokoko (free), and the engine-side retargeters (UE IK Retargeter, Unity Humanoid).

NLA and pose library

  • NLA: push actions down as strips (track.strips.new(name, start, action)) to layer/blend clips (blend type REPLACE/ADD/COMBINE), and use Tweak mode to edit. Strips need an action slot set when the action has several slots (strip.action_slot).
  • The pose library (bundled pose_library) stores poses as assets: poselib.create_pose_asset(pose_name=...), poselib.apply_pose_asset / blend_pose_asset. Use it for hand shapes, face poses, and a T/A rest pose.

Shape keys and other properties

Any property can be keyed: kb.keyframe_insert("value", frame=f) (shape key), mat.node_tree.nodes[...].inputs[...].keyframe_insert("default_value") (materials, which engines usually ignore), custom props ob.keyframe_insert('["prop"]').

Gotchas

  • pb.matrix = ... followed by keyframe_insert needs the right scene.frame_set(f) first, or an evaluation overwrites the pose.
  • Connected bones (use_connect) can't translate. Hips/pelvis must be unconnected from the root.
  • bake_action with visual keying reads evaluated poses, so constraints, IK and drivers are all captured. Mute the target's own constraints afterward (the script does), or they keep overriding the keys.
  • Quaternion vs Euler: game bones should use QUATERNION (no gimbal lock, smooth slerp). Euler is fine for single-axis controls. Don't mix modes between keys.
  • glTF sampling: export_force_sampling bakes every frame (the safe default). Non-linear F-curve modifiers and drivers only export when sampled.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.