blender-character-workflow

Blender Character Workflow

Read the shared execution guidance once per task for autonomous decisions, evidence cadence, and long-task continuity.

Match completion to the requested character: a static sculpt needs form and surface evidence; a rigged or animated character also needs deformation evidence. A mesh plus an armature is not a finished rig. Clarify static versus rigged delivery if the request and context leave that consequential choice open.

Establish the character contract

Record intended platform and export format, body/face scope, scale and axis, triangle/material/texture budgets, bone naming and hierarchy requirements, required controls, facial shape keys, poses/actions, and engine/avatar constraints. For a requested rig, identify the minimum acceptance poses: neutral, extreme bend for each major joint, reach, twist, locomotion/contact, and any required facial expression.

Review character form and fit

Read character form and fit review for humanoid creation and substantial anatomy/clothing repairs. Establish torso, pelvis, boot and hand proportions against the references before adding detail. Inspect continuous garment/joint surfaces, fitted accessories and skin-texture consistency in neutral and moving poses. These are visual acceptance criteria; passing rig/export metrics alone does not satisfy them.

For game characters, resolve silhouette, face identity and fitted garment surfaces before accessory detail. Inspect hands, boots, back and profile at the intended gameplay distance; adding seams to box-like anatomy does not finish it. For a rigged brief, test bend, reach and twist while forms are still editable, then repeat affected poses after clothing/accessory changes. Reserve time for this review and fresh-export deformation rather than spending the whole budget on a neutral hero pose. Static briefs need no invented rig.

Build and rig deliberately

For static character requests, build and review the form without inventing a rigging requirement. The rigging and motion steps below apply when requested.

Retrieve anatomy references with compatible species, age and pose before building a new character or creature. For a low-poly brown bear, for example, look for whole-body front, profile and three-quarter views in a comparable stance, plus head and paw close-ups. Record body/head proportions, muzzle depth, shoulder shape and foot contact from those images. Use separate low-poly examples for plane simplification; do not replace anatomy with a generic animal silhouette or infer all four legs from one occluded photograph.

1. Keep the authored mesh, armature, controls, deformation helpers, and export mesh in named collections. Apply or deliberately preserve transforms before skinning; document the choice.
2. Create anatomically or mechanically plausible joint placement and semantic bone names. Separate deform bones from animator controls where the target benefits from it.
3. Keep symmetry and mirror workflows explicit. Limit vertex influences and normalize weights according to the target runtime requirements.
4. Add IK/FK, constraints, corrective shapes, or drivers only when they solve a named deformation need. Test without relying on a single flattering pose.
5. For facial work, name shape keys semantically and test combinations that reveal volume loss, collisions, or eyelid/mouth failures.

Validate export and motion

1. Render the neutral and every contract pose from front, profile, back, and close-up deformation views. Open the evidence.
2. Check volume preservation, joint collapse, candy-wrapper twisting, clipping, foot/hand contact, constraint cycles, control readability, and unwanted scale/shear.
3. Export to the requested format, fresh-import it in a clean Blender process, and compare armature hierarchy, skin weights where supported, materials, actions, frame ranges, scale, and orientation.
4. Use $blender-agent-studio:blender-animation-workflow for action timing and $blender-agent-studio:blender-asset-validation for authored/exported geometry and hierarchy evidence.

Fit modular VRChat hair

For arbitrary head and hair assets, use scripts/fit_vrchat_hair.py. It fits the largest semantic hair mesh against the upper cranial region in evaluated world space while applying one uniform transform to the complete hair rig. Do not align the full hairstyle minimum Z to the head top; hanging strands and ponytails make that measurement invalid.

Require a 1-8 mm median fitted clearance, no more than 15 mm p95 clearance, fewer than 2% large gaps over 20 mm, and a 1.02-1.10 hair-to-cranium width ratio. Preserve hair armatures, bones, weights, materials, UVs, and separately named style meshes. Treat nearest-normal sign as a warning on multi-component or open head meshes, not proof of penetration; confirm intersections in fixed front, back, side, top, and perspective evidence. A Blender style visibility setup is not proof that Unity/VRChat expression-menu toggles were authored.

Completion gate

Resolve or explicitly report the form/fit review findings. Include full-body and feature close-ups from the final revision, with affected poses rechecked after the last geometry/material edit.

Deliver the editable source, requested export and fresh-import evidence. For a rigged/animated character, include the bone/control map and tested pose/action list. State any target-specific validation that could not be performed in Blender; a successful GLB import is not proof of runtime avatar compatibility.