Expert retopology is a plan executed with the fewest loops that capture the forms and let the mesh deform: loops close around eyes, mouth and joints, poles sit where nothing moves, and the subdivided surface, not the cage, matches the sculpt. An agent without a mouse can build every piece of this from data, but it reaches expert quality only on the parts it plans explicitly. Automatic remeshing is a finish for static assets and a start for everything else. If a sibling skill named here is missing from your available skills, ask the user to install it (npx skills add scenario-labs/skills --skill <name>); unattended, proceed from tool schemas and flag the gap.
REQUIRED BACKGROUND: scenario-blender-expert (execution channel, review loop, 5.2 API changes). Toolkit: scripts/bx_retopo.py (import sys; sys.path.append(\"<skills>/scenario-blender-retopology/scripts\"); import bx_retopo as R), all coordinates in the sculpt's local space.
R.surface_point), then builds.R.fit_subdiv. Kaspar's Multires Apply Base does the same job but overshoots (measured below).| Input | \nWhy it changes the plan | \nDefault when silent | \n
|---|---|---|
| Purpose | \nfilm cage renders at subdivision 2 (Kaspar); game ships triangles and bakes | \ndeforming character: film-style all-quad cage | \n
| Budget | \nfilm: no vertex budget, minimum loops (Kaspar); game: triangle budget | \nfilm stylized head 1,650 to 4,100 faces, body 9,400 to 16,500 (Blender Studio meshes); game character < 10,000 tris | \n
| Deformation | \nface rig, joints, how far the mouth travels (expression bounds) | \nface + limbs deform | \n
| Camera | \ndetail and density follow the closest view (SpeedChar) | \nclose-up face | \n
| Style | \nstylized: creases in topology; realistic: generic base + displacement (Kaspar) | \nstylized | \n
| Symmetry | \nR.topology_report(sculpt)[\"sym_local_pct\"], local space | \nMirror only above ~98 % (the sheep is 21 %, posed) | \n
| Separate parts | \neyes, teeth, hair clumps, clothing (from the body duplicate) | \nseparate objects | \n
Inspect (procedures P1): holes, manifoldness, scale, rotation, symmetry. GATE: all four written down.
\nMap: read the sculptor's face sets first (R.face_set_map, R.face_set_review, P1b: lip line, eye masks, ears, hooves); landmarks with coordinates, loop counts per feature from the Numbers table, crease list (check expression shape keys), per-part density. GATE: map exists before any geometry.
Choose a method per part:
\n| Part | \nMethod | \nEvidence | \n
|---|---|---|
| Eyes, mouth, brows (deforming) | \ncarve_rings(corners=, count=(lo, hi)): upper = lower by construction, geodesic rings, the border's count kept when inside the range; pair odd boundaries first (parity_strip) | \nv2: eyes 24 (12/12), mouth 30 (15/15) on every clean ring; forcing Kaspar's 18/24 cost 40 more poles | \n
| Nasolabial, face frame, neck (map lines) | \ncarve_band along enclosing_contour of face sets (P7): one clean loop, poles pushed onto the base | \nv2: nasolabial +18 poles; face frame opt-in, +64 poles on the sheep | \n
| Limbs, ears, tail, fingers, horns | \nlimb_profile (geodesic contours, root and joints found) + socket_tube(count=) (P3b, P8); straight parts: tube from sections | \nv2: all 7 parts incl. the 1 m curled tail, 3 rings per joint | \n
| Skull, torso, smooth masses | \nQuadriFlow start acceptable, else grid patches | \nlittle deformation | \n
| Static prop, scan, bake-only | \nQuadriFlow whole object (R.quadriflow) | \nLampel, Grant Abbitt | \n
| Realistic human head | \nshrinkwrap a clean generic base mesh, then fit (Kaspar BCON22 [00:20:16]) | \nnot scripted here | \n
Build islands separately and roughly (Kaspar: \"don't focus on tweaking all the points\"). Stack via R.setup_cage (Mirror clip > Shrinkwrap Target Normal Project > Subsurf hidden in edit mode). GATE: topology gate below passes per island.
Relax and freeze hand-built layouts: R.relax, R.freeze. Not after QuadriFlow (worsened coverage 0.44 % to 0.55 %).
Connect by counts: R.boundary_loops, then R.bridge / R.weld_loops on equal counts; reduce the larger side first (Kaspar: wrist 10 vs hand 16 was \"six too many\"). GATE: no triangles created.
Volume: film R.fit_subdiv (gentle defaults); game R.push_outside then R.triangulate_twisted and a Triangulate modifier on export.
Review and hand off: gates, two render sheets, handoff hygiene (P13). Report what the agent did not plan.
\ntests/code/blender-retopology/sheep_results.json and compare/metrics.json)| Strategy | \nFaces | \nSubdiv-2 mean error | \nInside | \nEye / mouth rings | \nClosed-loop edges | \nLooks | \n
|---|---|---|---|---|---|---|
| Kaspar's own retopo | \n2,596 | \n0.24 % | \n47 % | \n5,5 / 2 | \n29.9 % | \nclean, reads unsubdivided | \n
| QuadriFlow, area-matched budget | \n4,372 | \n0.14 % | \n87 % | \n0 / 0 | \n1.6 % | \nfaceted, torn ear, uniform density | \n
+ gentle fit_subdiv | \n4,372 | \n0.045 % | \n53 % | \n0 / 0 | \n1.6 % | \nsmooth masses, lumpy hooves | \n
| + Multires Apply Base instead | \n4,372 | \n0.11 % | \n15 % | \n0 / 0 | \n1.6 % | \novershoot | \n
| Hybrid: eye rings + ear tubes + gentle fit | \n4,544 | \n0.042 % | \n52 % | \n3,3 / 0 | \n5.9 % | \nfirst skill version | \n
| E1 run with the skill (fresh agent: face sets, warp, 4 eye rings, 5 lip rings, limb tubes) | \n5,778 | \n0.040 % | \n51 % | \n4,4 / 5 (46 to 48 verts) | \n19.4 % | \nplanned face, rippled lids, lumps at leg joins | \n
v2 tools, final (test_sheep_v2.py: face sets, counts as ranges, corner split, geodesic rings and tubes, nasolabial loop) | \n4,980 | \n0.028 % | \n52 % | \n4,4 clean 24-vert rings, 12/12 / 5 x 30, 15/15 | \n33.9 % | \nbest agent result: poles 5.7 % (Kaspar 5.2 %), none on clean rings, no 6-poles; nose and upper-lip patch irregular (human pass) | \n
Numbers flatter the agent: an aggressive fit scored 0.043 % and looked lumpier than Kaspar's cage. What stays human-level only: articulation counts with handle loops, crease loops placed from expression shapes, pole routing and loop budgets out of the face. Say so in the report.
\n| Item | \nValue | \nSource | \n
|---|---|---|
| Render subdivision (film) | \n2 levels; viewport 1 | \nKaspar Live #6 [00:21:53]; BCON22 frame 00:26:38 | \n
| Joint | \n3 loops over the bending part | \nKaspar Live #5 [01:37:48] | \n
| Limb start ring | \n8 verts (legs 8 then 12); Dikko arms 8 min, 10 to 12 typical, legs 18 | \nKaspar Live #4 [00:06:54]; Dikko [00:13:47] | \n
| Eyelid | \n2 loops per lid crease + ~4 across; 10/10 spans example | \nKaspar Live #1 [02:11:22]; Dikko [00:04:16] | \n
| Studio mesh | \n99.9 %+ quads, 0 to 4 valence 6+ (report and justify each), poles ~1.8 % head, ~5 % body, edge CV 0.33 to 0.8 | \ncalibration (bx_audit on Blender Studio files) | \n
| Auto-remesh hint (not a gate) | \nedge CV ~0.17 to 0.27 on an unedited uniform remesh, near 0 % closed loops | \ncalibration, sheep | \n
| Game | \nhead 2,000 to 3,000 tris, hands ~1,000 both, body 5,000 to 6,000 | \nSpeedChar ep23 [00:09:58], ep24 [00:50:34] | \n
| Apply Base levels | \n2 body, 3 head, hair, clothes | \nKaspar Live #5 [02:15:03] | \n
Code (thresholds [added], calibrated on Kaspar's sheep: 0.24 % mean, 47 % inside, coverage p95 0.82 %):
\nretopo_gate = R.topology_report(retopo)\nfit_gate = R.fidelity(retopo, sculpt, levels=2)\ngate = {\"six_poles_to_justify\": retopo_gate[\"valence_6plus\"], \"no_rim_poles\": retopo_gate[\"rim_poles\"] == 0,\n \"quads\": retopo_gate[\"non_quads_visible\"] == 0 or retopo_gate[\"quads_pct\"] >= 99.0,\n \"subd_fit\": fit_gate[\"mean_pct\"] <= 0.3 and 40 <= fit_gate[\"inside_pct\"] <= 60,\n \"coverage\": fit_gate[\"cov_p95_pct\"] <= 1.0, \"loops\": R.loop_stats(retopo)}\nprint(gate)\nEdge-length CV is never pass/fail (studio meshes read 0.33 to 0.8); 6-poles are reported and justified, not banned (studio 0 to 4). Also: crease_poles == 0, center_poles == 0 on mirrored faces, R.loops_around(retopo, eye, normal, max_radius) returns at least 3 closed rings per eye and mouth, ring counts equal before every merge, triangles only in hidden or rigid areas, budget met (R.tri_count). Visual (bx_review, same views as the sculpt): cage in wire (rings close, no wandering loops, poles in still areas) and subdivided matcap (lumps, pinching at poles, creases held), then the cage alone: it must read without subdivision (Kaspar Finale [01:58:41]). Score with references/critique.md.
| Mistake | \nLooks like | \nFix | \n
|---|---|---|
| Carving around QuadriFlow's closed eye slits | \nevery eye ring odd (47, 45), no all-quad fix | \nR.parity_strip between the two odd boundaries BEFORE carving | \n
| Counts inherited from a dense cut, or forced far below it | \neye rings 46 to 48; or pole density 9.5 % | \ncount=(lo, hi): each reduction unit costs 4 poles; ranges took the sheep from 9.5 to 5.7 % | \n
| 3D-nearest lookups where surfaces touch | \na band or socket cut bites an ear lying on the cheek | \ngeodesic fields, curve_verts, oriented nearest_vertex (built into bands and tubes) | \n
| Tight rings for an edge-flow loop | \na ridge after subdivision | \nband spacing = edge length; tight only for creases | \n
| Shipping a snapped cage | \nsubdivided surface inside the sculpt (87 %) | \nR.fit_subdiv | \n
| Shrinkwrap offset for volume | \ncavities worse | \nfit, offset 0 | \n
Raw quadriflow_remesh on a small dense mesh | \nCANCELED, \"needs to be manifold\" | \nR.quadriflow (scale trick, voxel fallback) | \n
| QuadriFlow symmetry left at its default True | \nwrong half on a posed mesh | \nsymmetry=False unless symmetric | \n
| Relaxing an auto-remesh | \ntips and ears shrink | \nrelax hand-built layouts only | \n
| Mark Sharp / Seam as reminders | \nshading breaks (4.1+), unwrap cuts | \nretopo_crease attribute (P5) | \n
| Grid fill on an odd loop | \nFINISHED, 0 faces | \ncount first; R.grid_fill raises | \n
| Connecting parts early | \nevery new loop runs everywhere | \nislands, count, merge last | \n
| Trusting mean error | \nlumpy fit passes | \nmatcap review, gentle fit | \n
| Re-shrinkwrap after Apply Base | \nvolume lost again | \nnever (Kaspar Live #6 [00:20:08]) | \n
| Game quads left non-planar | \nlines in the normal map | \nR.triangulate_twisted, Triangulate on export | \n
overlay.show_retopology, offset 0.01 m since 4.5); \"Project Individual Elements\" is snap_elements_individual = {'FACE_PROJECT'}. GUI only.bx_gui); headless use R.relax. F2, LoopTools, BSurfaces are extensions; mesh.circularize, space_edge_loops_evenly, flatten are built in; mesh.select_by_pole_count exists (4.4).sharp_edge is always honored since 4.1; Auto Smooth is gone (Smooth by Angle modifier).object.multires_base_apply(apply_heuristic=True); QuadriFlow lost its shortcut and has the traps above.references/procedures.md: tested code, P0 to P14 (inspect, face sets P1b, stack, tubes, geodesic limbs P3b, freeze/relax, annotations, QuadriFlow, map-line loops and rings, sockets, volume, checks, game, expression test, handoff, AI-mesh finishing). Load before writing any code.references/expert-notes.md: principles by expert with timestamps, face and body loop maps, clothing, game rules, the sheep experiment in full. Load when planning a face, body, clothing or game mesh.references/critique.md: the rubric to judge your own retopo. Load at every review.references/sources.md: all 24 source videos with best timestamps.