QUADIFY
By Anvil Interactive Solutions
← Home

Scenarios — Which Engine to Use

This page covers the most common mesh types you'll encounter and the exact recommended workflow for each. Every scenario is derived from the preset definitions, algorithm routing logic, and known failure modes in the codebase.

Vehicle / Car Body Panels

Mesh characteristics: smooth curved hard-surface panels, feature edges from booleans or CAD export, typically 2k–20k faces per panel, may have n-gons from arbitrary triangulation.

Recommended workflow:

  1. Apply the Vehicle / Car Body preset from the Tools panel.
  2. Use Quick Remesh: What is your mesh? → Hard surface / What do you need? → Best topology for curved panels, or Fast preview for hoods and windshields.
  3. If n-gons remain from CAD triangulation, try Dissolve + Requadify from the Tools panel before Quadify — it often recovers true panel topology in one pass.
  4. Post-process with Reduce Poles on any open panels.

Avoid: QuadWild on car panels — it doesn't preserve feature edges and produces worse loops than Smart or Hard Edge on this mesh type.

Expected result: 90–100% quads on flat panels, 85–92% on curved surfaces. If below 85%, try Hard Edge Preserving preset — it anchors loops to sharp edges more aggressively.

Organic / Characters / Creatures

Mesh characteristics: dense sculpted surface with no sharp edges, high genus (many holes/handles), typically needs full topology replacement rather than conversion.

Recommended workflow:

  1. Apply the Organic (Characters, Creatures) preset.
  2. Quick Remesh: What is your mesh? → Organic/character / What do you need? → Best topology (routes to QuadWild Fine).
  3. If the mesh has an armature attached, run Compute Animatability Score in Topology QA before and after to verify joint zone quality improved.
  4. Post-process problem zones with Local Cleanup in Edit Mode.

If QuadWild times out: the mesh is likely too dense. Run Reduce Poly Count from the Tools panel to bring it under 50k faces, then retry QuadWild.

If the mesh is high-genus (ornate armour, jewellery, lattice detail): QuadWild hangs on these. Switch to Exact face count in Quick Remesh (routes to Voxel + QuadriFlow) — it handles high genus without hanging.

Expected result: 95–100% pure quads with QuadWild Fine. QuadriFlow produces 90–98% quads depending on surface complexity.

Photogrammetry / 3D Scan Data

Mesh characteristics: extremely dense all-triangle mesh from photogrammetry (Metashape, RealityCapture, Meshroom) or structured-light scanner. May have non-manifold edges, scan noise, boundary artefacts, and no meaningful edge flow.

Recommended workflow:

  1. Run Scan Selected Mesh in the AI Mesh Report panel first — get a defect inventory before proceeding.
  2. Apply the Scanned Data preset.
  3. Quick Remesh: What is your mesh? → Scan data / What do you need? → Best topology for maximum quad quality, or Exact face count if you need a specific poly budget.
  4. Voxel Remesh (auto-selected for scan data) handles broken topology and noise that would cause Smart or QuadriFlow to fail.
  5. Check the AI Mesh Report after the operation for defects fixed and UV transfer status.

For very dense scans (500k+ faces): manually decimate to ~100k faces first using Reduce Poly Count, then run the scan workflow. Voxel + QuadriFlow above 500k may cause temporary Blender unresponsiveness.

Expected result: 90–98% quads, all non-manifold and degenerate geometry resolved. UV maps are destroyed by Voxel Remesh — run Auto UV re-projection or bake from original.

AI-Generated Meshes (Meshy, Tripo, Rodin, CSM, Luma)

Mesh characteristics: dense all-triangle output, typically 100k–500k faces, often with non-manifold edges from generation artefacts, no UV seams, no meaningful edge flow. Detection thresholds: ≥60% tris with NM≥5 and no seams, or ≥80% tris on meshes ≥200k faces.

Recommended workflow:

  1. Apply the AI Generated preset from the Tools panel.
  2. Quick Remesh: What is your mesh? → AI-generated / What do you need? → Best topology.
  3. The preset routes through a decimate-first pipeline (reduces to ~50k) then QuadWild MEDIUM. This avoids QuadWild timeouts on dense inputs.
  4. Check the AI Mesh Report panel — it shows every defect fixed and the before/after topology comparison.
  5. If the mesh has textures from the AI generator, run Bake Texture from Original after remeshing to recover the appearance on the new topology.

If the mesh isn't auto-detected as AI-gen: apply the preset manually — it will still route correctly regardless of detection.

Expected result: 90–100% quads after QuadWild, depending on surface complexity. Generation artefacts (non-manifold edges, zero-area faces) are resolved during the intake pipeline.

CAD Imports / Boolean Results

Mesh characteristics: flat panels with many n-gons from tessellation, coplanar edges from face subdivision, hard feature edges from boolean cuts, inconsistent triangulation direction.

Recommended workflow:

  1. Apply the Hard Surface preset.
  2. Try Dissolve + Requadify first (Tools panel) — for CAD meshes this often produces perfect results in one pass by recovering the underlying panel geometry.
  3. If Dissolve doesn't fully clean it up, run Quick Remesh: Hard surface / Best topology (routes to Hard Edge Preserving + Smart).
  4. Use Feature Analysis → Gaussian Curvature to visualise where boolean cuts and feature edges are before choosing edge handling settings.

Expected result: 95–100% quads on flat surfaces after Dissolve. Curved CAD surfaces may need a second pass with the Hard Edge preset.

High-Genus / Lattice / Ironwork / Filigree

Mesh characteristics: many holes, thin struts, complex topology where QuadWild's cross-field decomposition cannot converge. QuadWild hangs indefinitely on these regardless of face count.

Recommended workflow:

  1. Do not use Best topology (QuadWild) for these mesh types.
  2. Quick Remesh: What is your mesh? → any / What do you need? → Exact face count. This routes to Voxel + QuadriFlow, which handles high genus without hanging.
  3. Set the target face count to match your LOD budget.
  4. After remeshing, voxelisation may merge thin struts — inspect the result and adjust Voxel Detail Level in Algorithm Settings if too much detail is lost.

Expected result: 90–100% quads. Surface detail of very thin struts may be approximated by the voxel step — increase Voxel Detail Level for finer preservation.

Matching a Reference Topology

Goal: apply the topological character of a hand-retopologised or cleaned reference mesh to a new target (e.g. apply the same door panel topology to three scan variants of the same door).

Recommended workflow:

  1. Open the Reference-Driven Retopo sub-panel.
  2. Pick the reference mesh — must be at least 60% quads for useful results.
  3. Choose mode: QuadriFlow for clean targets, Voxel for noisy/damaged targets, Smart only if the target already has near-quad topology.
  4. Click Apply Reference Topology.

Expected result: target topology density and edge flow approximate the reference. Exact pole positions will differ — this is a density and flow guide, not a vertex-for-vertex match.

Batch Processing a Scene

Goal: process a full scene of mixed mesh types (vehicle, props, environment assets) in one pass.

Recommended workflow:

  1. In Object Mode, shift-click to select all mesh objects, or press A to select all.
  2. Open the Batch Processing sub-panel — it shows the count of selected mesh objects.
  3. Click Batch Quadify. Ultra processes each object sequentially, running ML routing per object.
  4. After completion, the batch results table shows per-object: algorithm used, quad %, pass/fail against threshold.
  5. Objects below threshold are flagged with ⚠ — click them individually to run Try Alternative.
  6. Click Export Stats CSV for a full audit log.

For large batches (20+ objects): run a test batch of 3–5 representative meshes first to verify ML routing before committing the full scene. This also populates the experience DB for better routing on the full run.

Pre-rig Topology QA

Goal: verify topology is deformation-ready before handing a mesh to a rigger.

Recommended workflow:

  1. Open Topology QA sub-panel → click Full Topology QA.
  2. View the heatmap: switch Viewport Shading → Color → Vertex. Green = clean, yellow = minor issues, red = poles or stretched quads.
  3. Click Compute Animatability Score. If an armature is attached, it uses bone positions for joint zone analysis. Otherwise it estimates from curvature.
  4. Review the zone breakdown — each joint zone gets a 0–100 score. Focus on anything below 55.
  5. Use Export HTML Report to share the analysis with the rigger.
  6. For red zones: switch to Edit Mode, use Select N-gons or manual selection, then run Local Cleanup or Fix N-gons on those faces.
  7. Re-run Animatability Score to confirm improvement.