Pipeline Overview: 7 Steps

CAD Source → Format Conversion → Retopology → UV Unwrap → Channel Split → Texture Compress → glTF Pack → Ship
    ①            ②                ③           ④           ⑤              ⑥             ⑦

Every step has an explicit pass/fail threshold. Skip any and all downstream optimizations get diluted.

① CAD Source Acceptance

Ranked by suitability (best → worst):

Format Preserves Topology Preserves Assembly Preserves Material Names Grade
STEP / Parasolid (.x_t) ✅ ✅ ⚠️ ⭐⭐⭐⭐⭐
SolidWorks .sldasm ✅ ✅ ✅ ⭐⭐⭐⭐
Rhino .3dm ⚠️ ⚠️ ⚠️ ⭐⭐⭐
FBX / OBJ (render exports) ❌ (over-tessellated) ❌ ⚠️ ⭐⭐ last resort only
STL ❌ ❌ ❌ ⭐ NEVER USE

Acceptance checklist:

  • Part names are human-readable (e.g. FAB-001-Leg-Left — not Mesh.287)
  • Assembly hierarchy is clean (≤ 5 sub-assembly levels deep)
  • No part interpenetrations (causes AO baking artifacts)

② Format Conversion: Preserve Assembly Hierarchy

Most common mistake right here: people dump CAD straight into Blender → export OBJ → lose the entire assembly tree → spend 10× engineering effort later re-splitting individual parts for the configurator.

Correct toolchain, ranked:

  • Datakit CrossManager (industry gold — 99% format coverage)
  • Autodesk Forge (cloud API, great for batch workflows)
  • ANSYS SpaceClaim (indispensable for healing CAD geometry issues)

Output target: An intermediate .fbx (or equivalent) that retains part naming + assembly hierarchy — zero decimation at this stage.

③ Retopology & Decimation: Triangle Budget

Category Target tris per SKU Mobile target
Chairs / small furniture 8,000 – 15,000 ≤ 12,000
Sofas / modular seating 20,000 – 40,000 ≤ 30,000
Doors / windows / arch parts 15,000 – 35,000 ≤ 25,000
Small consumer electronics 6,000 – 12,000 ≤ 8,000
Industrial machinery (lite) 50,000 – 100,000 ≤ 70,000

Decimation tools ranked:

  1. Simplygon (best-in-class automated; protects UVs & seams)
  2. InstaLOD (price-friendly, near-Simplygon quality)
  3. Blender Decimate modifier (manual fallback, not recommended for batch)

⚠️ Hard constraint: No visible normal tearing after decimation. If it exists → go back to the original CAD and heal the geometry first. Don’t try to patch it in decimation.

④ UV Unwrap: You Need (at least) 2 UV Sets

PBR pipelines require a minimum of 2 UV sets — we recommend 3:

UV Set Purpose Rules
UV0 BaseColor / Normal / Roughness / Metallic Zero overlap in 0–1 space, ≥ 85% utilization
UV1 Ambient Occlusion / Lightmap Overlap OK by bake policy; keep 4–8px gutter between charts
UV2 (Optional) Decals / custom engraving Per-face, as needed

Best-in-class tool: RizomUV (fastest auto-symmetric unwrap by a wide margin).

⑤ Channel Split: Only 4 Textures Matter for Web

Web PBR ≠ offline render PBR. Don’t export all 8 Substance maps — only 4 are required for production WebGL:

Map Format Compression Typical Size
BaseColor (+ Alpha if needed) PNG → KTX2 UASTC + RDO 1024² or 2048²
Normal (OpenGL convention) PNG → KTX2 UASTC Normal mode same as above
ORM (Occlusion-Roughness-Metallic packed) PNG → KTX2 R=AO, G=Rough, B=Metal same as above
Emissive (if applicable) PNG → WebP Lossy 85% same as above; delete if unused

⚠️ Common gotcha: Roughness MUST be gamma-corrected correctly. If your roughness map looks too dark, 9 times out of 10 it’s an sRGB/linear tag mismatch.

⑥ Texture Compression: KTX2 or Bust

It’s 2026 — stop shipping raw PNG / JPG textures:

Format Size ratio (quality-matched) iOS Android Desktop Chrome
PNG 100% (baseline) ✅ ✅ ✅
WebP ~50% ✅ 14+ ✅ ✅
KTX2 (UASTC) ~25% ✅ 16+ ✅ 12+ ✅ 106+
KTX2 (Basis ETC1S) ~15% ✅ ✅ ✅

Compression tool: Khronos’ official toktx CLI, or our online batch compressor.

Hard target per SKU: All assets (geometry + textures) combined ≤ 3MB first-screen; ≤ 2MB on mobile.

⑦ glTF Pack & Deploy

Final pass with glTF Transform — 3 operations:

  1. Inline all textures into a single .glb (prevents request waterfalls)
  2. Draco-compress geometry (~70% typical compression)
  3. Batch-convert all remaining PNG textures → KTX2
// Minimal glTF Transform snippet
import { NodeIO } from '@gltf-transform/core';
import { KHRONOS_EXTENSIONS } from '@gltf-transform/extensions';
import { draco, textureCompress } from '@gltf-transform/functions';

const io = new NodeIO().registerExtensions(KHRONOS_EXTENSIONS);
const document = await io.read('input.glb');

await document.transform(
  draco({ method: 'edgebreaker' }),
  textureCompress({ encoder: 'ktx2' })
);

await io.write('output.glb', document);

Final 4-Point Gate Before Ship

Every SKU must pass before going live.

Metric Threshold Measurement
LCP < 1.5s (4G) Chrome Lighthouse
FPS ≥ 60 desktop / ≥ 45 mid-tier mobile Manual spot-check via renderer.info
Peak RAM ≤ 250 MB (mobile) Chrome DevTools Memory
Texture fidelity No visible blocking at 200% zoom Visual sign-off + side-by-side

Fail any gate → go back to the relevant step. Do not ship broken assets.

Closing

A high-quality asset pipeline is the bedrock of a configurator’s perceived quality. Invest a week standardizing your pipeline now — it pays for itself 10× over in firefighting avoided later.