Imported from dodontommy/darkfantasy (skills/blender-rig-pose-director/SKILL.md). Install upstream with npx skills add dodontommy/darkfantasy --skill blender-rig-pose-director. Copyright stays with the author.

Blender Rig & Pose Director

When to invoke

• Body + costume topology is finalized (post-retopo, post-character-topology-reviewer) and a rig does not yet exist.
• A ticket asks for skinning, weight painting, IK setup, or pose-key authoring.
• Cape or skirt needs secondary motion (sim, bone chain, or hybrid).
• A render/export ticket is blocked because the armature is missing the noble-contrapposto pose.
• Choosing between Rigify, Auto-Rig Pro, and AccuRIG 2 for a new character variant.

Role

Owns everything from "the mesh is sealed" to "the armature is bind-posed, weighted, posed, and ready for sim-bake or render". Delegates geometry edits back to the asset-builder skills; delegates final shader bakes back to blender-shader-builder. Encodes the L2 noble-fidelity bar (~180 deform bones, severe contrapposto, weapon/spell-focus hand) and the headless-Rigify recipe.

This skill owns rig taste. The pose is half the silhouette — wrong pose, dead character. The Matriarch is not in T-pose for any deliverable; she is in a 3/4-low-angle severe-contrapposto noble stance from the moment frame 1 exists.

Rig generator decision matrix

Source: docs/research/dark-fantasy-shading-pipeline.md §E1.

Recommendation for the Nocturne Matriarch: Rigify body + custom shape-key face panel (subtle noble expressions read better than Rigify's bone face for this character). Add finger metarig for the clawed gauntlets. Expect ~180 deform bones after generation: ~60 body + 30 face shape-key drivers (counted as bones for budget) + 8 cape + 12 skirt + 12 hair + 8 pauldron secondaries.

Rigify metarig customization for the Matriarch

Source: docs/research/lineage2-art-style.md § "Direct Guidance" (proportions) + docs/research/headless-blender-2026.md §7.

Customizations applied to bpy.ops.object.armature_human_metarig_add() output:

1. Proportion lock to 7.8 heads. Scale the metarig so head-bone length × 7.8 = total height from heel to crown. Adjust spine, spine.001..006, and leg chain proportionally.
2. Narrow shoulders. Move shoulder.L / shoulder.R head positions inward so the shoulder-tip-to-shoulder-tip distance equals 1.7 × head-width (Korean-Gothic-couture silhouette).
3. Add finger metarig for clawed gauntlets — the default human metarig already includes f_index/middle/ring/pinky plus thumb chains, but verify all 4 phalanges per finger are present (3 phalanx bones + palm). Generate clawed pose-shape later via pose-mode rotation, not topology.
4. Eye bones — keep the default eye.L / eye.R; they drive look-at via the eyes controller, useful for portrait-render eye contact.
5. Pelvis cock prep — leave hips neutral in the metarig; the contrapposto rotation is applied at the generated rig, not baked into bind pose.

Headless Rigify generation

Full recipe — paste into scripts/parts/rig_generate.py:

import bpy
from mathutils import Vector

# 0. Clean slate, factory startup assumed (--factory-startup)
bpy.ops.preferences.addon_enable(module="rigify")

# 1. Add human metarig
bpy.ops.object.armature_human_metarig_add()
metarig = bpy.context.object
metarig.name = "matriarch_metarig"

# 2. Lock 7.8-head proportion. Head bone length is the unit.
arm = metarig.data
bpy.ops.object.mode_set(mode='EDIT')
head_bone = arm.edit_bones["spine.006"]  # the head bone in the human metarig
head_len = (head_bone.tail - head_bone.head).length
target_total = head_len * 7.8
heel = arm.edit_bones["heel.02.L"].head.z
crown = arm.edit_bones["spine.006"].tail.z
scale = target_total / (crown - heel)
for eb in arm.edit_bones:
    eb.head *= scale
    eb.tail *= scale

# 3. Narrow shoulders to 1.7 head-widths.
head_width = head_len * 0.7  # rough head-width estimate
target_shoulder_span = 1.7 * head_width
for side in ("L", "R"):
    sb = arm.edit_bones[f"shoulder.{side}"]
    sign = 1 if side == "L" else -1
    sb.head.x = sign * (target_shoulder_span / 2 - 0.05)

bpy.ops.object.mode_set(mode='OBJECT')

# 4. Generate the rig
with bpy.context.temp_override(active_object=metarig, object=metarig,
                               selected_objects=[metarig],
                               selected_editable_objects=[metarig]):
    bpy.ops.pose.rigify_generate()

rig = bpy.data.objects["rig"]
rig.name = "matriarch_rig"

# 5. Verify deform bone count is in the ~180 envelope (body+face+secondaries
#    will be added later; expect ~120 from the base human rigify generation).
deform_bones = [b for b in rig.data.bones if b.use_deform]
assert 100 <= len(deform_bones) <= 140, f"Unexpected deform bone count: {len(deform_bones)}"
print(f"Rigify generated {len(deform_bones)} deform bones (body+face+fingers).")

Notes from docs/research/headless-blender-2026.md §7:

• rigify_generate runs in --background because all its operators have data-API paths.
• In 4.1+ the generated rig uses bone collections — use armature.collections_all["FK"].is_visible = True, not the removed armature.layers[].
• Custom shape bones live in ~/Library/Bone Shapes by default; for the Matriarch keep that naming so the rig UI panel renders correctly.

Cape physics — baked Cycles cloth

Source: docs/research/dark-fantasy-shading-pipeline.md §E2.

Cape = ceremonial heavy fabric. Settings for the cloth modifier on the cape mesh:

Bake to alembic for headless render so the sim is deterministic and parallel-safe across render workers:

# After cape sim is set up
bpy.ops.object.select_all(action='DESELECT')
cape = bpy.data.objects["cape"]
cape.select_set(True)
bpy.context.view_layer.objects.active = cape

# Bake cloth cache
override = {"scene": bpy.context.scene,
            "active_object": cape,
            "point_cache": cape.modifiers["Cloth"].point_cache}
with bpy.context.temp_override(**override):
    bpy.ops.ptcache.bake(bake=True)

# Export to alembic for headless re-import
bpy.ops.wm.alembic_export(
    filepath="//outputs/sim/cape.abc",
    selected=True,
    start=1, end=120,
    sh_open=0.0, sh_close=0.5,
    apply_subdiv=True,
)

Skirt physics — bone chain + low-poly-cloth-as-bone-target

Skirt = 6 panels per the costume order in docs/research/dark-fantasy-shading-pipeline.md §A4. Each panel gets a 5–8 bone chain, parented to the hip bone.

Three options, ranked:

1. Bone Dynamics Pro (paid, Blender Market) — drives bone rotations from spring/damper params; clean UI; deterministic. Recommended if budget allows.
2. Goo Physics (paid, Superhive) — purpose-built for cape/skirt secondary motion, stylized "anime" feel. Pick if the look needs more swing than realism.
3. Free DIY: low-poly-cloth-as-bone-target trick. Sim a 6×8 vertex grid per panel with the cloth modifier (cheap, real-time bakeable), then add a Copy Location constraint on each chain bone targeting its nearest cloth vertex. This gives cinematic movement at game-rig cost.

Snippet for the DIY trick:

# For each skirt panel: create low-poly cloth proxy, bind bone chain to it
import bpy

panel_names = [f"skirt_panel_{i:02d}" for i in range(6)]
chain_lengths = {n: 6 for n in panel_names}  # 6 bones per panel chain

for panel in panel_names:
    proxy = bpy.data.objects[f"{panel}_cloth_proxy"]  # 6×8 grid, cloth-simmed
    rig = bpy.data.objects["matriarch_rig"]
    bpy.context.view_layer.objects.active = rig
    bpy.ops.object.mode_set(mode='POSE')
    for i in range(chain_lengths[panel]):
        bone_name = f"{panel}_bone_{i:02d}"
        pb = rig.pose.bones[bone_name]
        # Snap to nearest proxy vertex via Copy Location
        c = pb.constraints.new(type='COPY_LOCATION')
        c.target = proxy
        c.subtarget = ""  # use whole-mesh nearest-vertex via vertex group
        c.influence = 0.85  # leave 15% bone authority
    bpy.ops.object.mode_set(mode='OBJECT')

Noble contrapposto pose preset

Source: docs/research/lineage2-art-style.md §9.

The canonical pose for the Matriarch's hero render. Apply at frame 1 (frame 0 reserved for bind pose).

Snippet:

import bpy
from math import radians

rig = bpy.data.objects["matriarch_rig"]
bpy.context.view_layer.objects.active = rig
bpy.ops.object.mode_set(mode='POSE')

scene = bpy.context.scene
scene.frame_set(0)
# Bind pose at frame 0 — keyframe identity transforms on every controller
for pb in rig.pose.bones:
    pb.keyframe_insert("location", frame=0)
    pb.keyframe_insert("rotation_euler", frame=0)
    pb.keyframe_insert("rotation_quaternion", frame=0)

scene.frame_set(1)

POSE = {
    "hips":       {"rotation_euler": (0, 0, radians(8))},
    "shoulder.L": {"rotation_euler": (0, 0, radians(-3))},
    "shoulder.R": {"rotation_euler": (0, 0, radians(3))},
    "spine.001":  {"rotation_euler": (0, 0, radians(-4))},
    "spine.003":  {"rotation_euler": (0, 0, radians(2))},
    "head":       {"rotation_euler": (radians(-3), 0, radians(-2))},
    "hand_ik.L":  {"location": (-0.4, -0.2, 1.0)},
    "hand_ik.R":  {"location": (0.35, -0.05, 0.45)},
    "foot_ik.L":  {"location": (-0.18, 0.05, 0)},
    "foot_ik.R":  {"location": (0.22, -0.08, 0)},
}

for bone, transforms in POSE.items():
    pb = rig.pose.bones[bone]
    for attr, val in transforms.items():
        setattr(pb, attr, val)
        pb.keyframe_insert(attr, frame=1)

# Add focus empty + damped track on the casting hand
focus = bpy.data.objects.get("focus_empty") or bpy.data.objects.new(
    "focus_empty", None)
focus.location = (0, -0.3, 1.1)
if focus.name not in bpy.context.scene.collection.objects:
    bpy.context.scene.collection.objects.link(focus)

hand_l = rig.pose.bones["hand_ik.L"]
if "noble_focus_track" not in hand_l.constraints:
    c = hand_l.constraints.new(type='DAMPED_TRACK')
    c.name = "noble_focus_track"
    c.target = focus
    c.track_axis = 'TRACK_NEGATIVE_Z'

bpy.ops.object.mode_set(mode='OBJECT')

# Markers for alternate poses
scene.timeline_markers.new("bind",          frame=0)
scene.timeline_markers.new("noble_primary", frame=1)
scene.timeline_markers.new("spell_cast",    frame=10)
scene.timeline_markers.new("blade_drawn",   frame=20)
scene.timeline_markers.new("crown_lift",    frame=30)

Pose key conventions:

• Frame 0: bind pose (identity). .blend saves with armature in pose-mode showing this.
• Frame 1: noble_primary — the canonical hero render pose.
• Frame 10: spell_cast — left hand raised, palm open, focus_empty at z=1.6.
• Frame 20: blade_drawn — right hand on pommel pulled outward, left hand at guard.
• Frame 30: crown_lift — both hands at temples, head tilted up 8°. (Variant for promo.)

Each marker has a corresponding render preset in blender-render-director.

Camera framing for portrait render

Source: docs/research/lineage2-art-style.md §9.

Set the hero camera (also exposed by blender-render-director):

• Lens: 70mm (within the 50–85mm L2 range; 70mm gives the cleanest noble-portrait compression).
• Camera height: 15° below eye line — eye-line is head bone tail z; place camera at head.z - tan(15°) × distance.
• Camera distance: 2.5–3.0 m from subject for full-body, 1.4 m for chest-up.
• Subject occupies central 60% of frame; cape and hair break the rule-of-thirds.

cam = bpy.data.objects["hero_cam"]
cam.data.lens = 70
# Place at eye line, then drop 15° angularly, then aim at sternum
import math
eye_z = rig.pose.bones["spine.006"].tail.z + rig.location.z
distance = 2.7
cam.location = (0, distance, eye_z - distance * math.tan(math.radians(15)))
# Track-to constraint on sternum empty
sternum = bpy.data.objects.get("sternum_empty")
tt = cam.constraints.get("hero_track") or cam.constraints.new(type='TRACK_TO')
tt.name = "hero_track"
tt.target = sternum
tt.track_axis = 'TRACK_NEGATIVE_Z'
tt.up_axis = 'UP_Y'

Quality gates

Checklist — every gate must be green before this skill's ticket closes:

1. No zero-weight verts on deform bones. Run bpy.ops.paint.weight_from_bones() only as fallback; manual weight cleanup preferred. Verify with:

   for v in mesh.data.vertices:
       total = sum(g.weight for g in v.groups if mesh.vertex_groups[g.group].name in deform_bone_names)
       assert total > 0.001, f"Vertex {v.index} has no deform weight"

2. Root bone at world origin. rig.pose.bones["root"].head_local == Vector((0,0,0)). The character's world origin is between the feet at z=0 (per character-export-validator rule 3).
3. Bind pose is the saved state. When the .blend is saved, the armature is in pose-mode with frame 0 active — anyone opening the file sees identity transforms, not the hero pose.
4. Deform bone count in envelope. 160–200 deform bones (~180 target). Anything outside means the metarig customization or the secondary-bone setup is wrong.
5. All custom shape bones use ~/Library/Bone Shapes naming convention. Verify with bpy.data.objects["WGT-*"] — the rig UI panel breaks if these are renamed.
6. Pose markers exist: bind, noble_primary, spell_cast, blade_drawn, crown_lift.
7. Cape sim baked to alembic at outputs/sim/cape.abc. Headless renders re-import this — they do not re-sim.
8. Skirt bone chains parented to hips, not to root. Hip rotation must drive skirt; root drives the whole character.
9. Promotion block emitted. Per orchestrator/README.md § "Source-of-truth promotion" — every MCP rig session ends with a ## Promotion block of the bpy operations, ready to paste into scripts/parts/rig_generate.py.

References

• docs/research/dark-fantasy-shading-pipeline.md §E1 (rig matrix), §E2 (cape/skirt physics).
• docs/research/headless-blender-2026.md §7 (Rigify in headless), §3.3 (temp_override), §6.1 (addon enable).
• docs/research/lineage2-art-style.md § "Direct Guidance" (proportions, ~180 bones), §9 (severe contrapposto, 70mm at 15° below eye line).
• orchestrator/README.md § "Source-of-truth promotion".
• scripts/create_dark_fantasy_lady_blockout.py — exemplar for headless script structure (__main__, pathlib.Path from __file__, factory-startup-safe addon enables).
• Hand-off targets: blender-shader-builder (texture bake before export), blender-render-director (hero portrait render), character-export-validator (when handing rigged character to a downstream engine).