threejs-agents-model-optimizer

threejs-agents-model-optimizer

Quick Reference

Optimization Pipeline Overview

INPUT (.glb/.gltf)
  |
  v
[1. ASSESS] ── polygon count, texture sizes, file size, draw calls
  |
  v
[2. MESH OPTIMIZE] ── dedup, flatten, join, weld, simplify
  |
  v
[3. TEXTURE OPTIMIZE] ── resize, KTX2 compress, atlas
  |
  v
[4. COMPRESS] ── Draco mesh compression, quantize attributes
  |
  v
[5. VALIDATE] ── visual diff, file size check, runtime test
  |
  v
OUTPUT (optimized .glb)

Target File Size Budgets

Platform	Max File Size	Max Polygons	Max Texture Size
Mobile web	2 MB	50K triangles	1024x1024
Desktop web	5 MB	200K triangles	2048x2048
Desktop app	20 MB	500K triangles	4096x4096
Hero asset (single)	1 MB	30K triangles	1024x1024

Tool Selection

Tool	Best For	Install
gltf-transform	Full pipeline, scriptable, Node.js API	npm i @gltf-transform/cli
gltfpack	Fast one-shot compression	npm i -g gltfpack
Blender	Manual mesh editing, UV repacking	Blender 3.6+

Critical Warnings

NEVER skip the assessment step -- optimizing blindly wastes time and can produce worse results than the original.

NEVER apply Draco compression AND meshopt compression to the same file -- they are mutually exclusive. Choose ONE.

NEVER use KTX2 UASTC for diffuse color maps on mobile -- UASTC textures are 8-16 bytes/texel in VRAM. ALWAYS use ETC1S for color maps on mobile targets.

NEVER simplify meshes below 10% of original without visual validation -- aggressive simplification destroys silhouettes and UV mapping.

ALWAYS validate optimized models visually before shipping -- automated metrics cannot catch all visual artifacts.

ALWAYS keep the original unoptimized model in version control -- optimization is lossy and irreversible.

---

Step 1: Assessment

Before optimizing, ALWAYS gather these metrics:

# Using gltf-transform CLI
npx gltf-transform inspect model.glb

# Key output to check:
# - Mesh count and total triangle count
# - Texture count, dimensions, and format
# - Total file size (uncompressed)
# - Accessor count (indicates potential deduplication)
# - Animation track count

Assessment Decision Tree

File size > target budget?
├── YES: Textures > 50% of file size?
│   ├── YES → Start with texture optimization (Step 3)
│   └── NO → Start with mesh optimization (Step 2)
└── NO: Draw calls > 50?
    ├── YES → Merge meshes (join/flatten)
    └── NO → Only apply compression (Step 4)

Polygon Count Guidelines

Asset Type	Target Triangles	Notes
Background prop	100-500	Minimal detail
Mid-ground object	1K-5K	Visible but not hero
Hero/focus object	10K-50K	High detail, close-up
Character	15K-30K	With LOD chain
Full scene	100K-300K	All objects combined
Architectural (BIM)	200K-500K	Simplified from CAD

---

Step 2: Mesh Optimization

gltf-transform Mesh Pipeline

ALWAYS run these operations in this order:

# 1. Remove duplicate accessors and unused data
npx gltf-transform dedup input.glb deduped.glb

# 2. Flatten node hierarchy (removes empty nodes)
npx gltf-transform flatten deduped.glb flat.glb

# 3. Join meshes sharing the same material
npx gltf-transform join flat.glb joined.glb

# 4. Weld vertices (merge vertices within tolerance)
npx gltf-transform weld joined.glb welded.glb --tolerance 0.0001

# 5. Simplify mesh (reduce triangle count)
npx gltf-transform simplify welded.glb simplified.glb \
  --ratio 0.5 \
  --error 0.001

# 6. Remove unused resources
npx gltf-transform prune simplified.glb output.glb

Combined Pipeline (Single Command)

npx gltf-transform optimize input.glb output.glb \
  --compress draco \
  --texture-compress ktx2

Mesh Simplification Settings

Quality Level	Ratio	Error Tolerance	Use Case
Minimal	0.75	0.0005	Subtle reduction, preserves detail
Moderate	0.50	0.001	Balanced quality/size
Aggressive	0.25	0.005	LOD1/LOD2 generation
Extreme	0.10	0.01	LOD3, distant objects only

Weld Settings

Tolerance	Effect
0.0001	Conservative -- merges only nearly-identical vertices
0.001	Standard -- good for most models
0.01	Aggressive -- may cause visible seams on hard edges

---

Step 3: Texture Optimization

Texture Decision Tree

Texture type?
├── Color/Diffuse (baseColorTexture)
│   ├── Mobile → ETC1S (KTX2), max 1024px
│   └── Desktop → UASTC (KTX2), max 2048px
├── Normal map
│   └── ALWAYS UASTC (KTX2) — ETC1S causes visible artifacts on normals
├── ORM (occlusion/roughness/metallic)
│   └── ETC1S (KTX2) — perceptual quality less critical
├── Emissive
│   └── ETC1S (KTX2) — unless HDR, then keep as-is
└── HDR environment
    └── Keep as .hdr or .exr — do NOT KTX2 compress

KTX2 Compression Commands

# ETC1S: smaller file, lower quality (color maps, ORM)
npx gltf-transform ktx2 input.glb output.glb \
  --slots "baseColorTexture,emissiveTexture,occlusionTexture" \
  --filter "baseColorTexture=etc1s" \
  --quality 128

# UASTC: larger file, higher quality (normal maps)
npx gltf-transform ktx2 input.glb output.glb \
  --slots "normalTexture" \
  --filter "normalTexture=uastc"

KTX2: ETC1S vs UASTC

Property	ETC1S	UASTC
File size	Very small (6-8x smaller)	Moderate (2-4x smaller)
VRAM usage	Small	Large (8-16 bytes/texel)
Quality	Good for color	Near-lossless
Decode speed	Fast	Requires transcoding
Best for	Color maps, ORM, emissive	Normal maps, detail textures

Texture Resize

# Resize all textures to max 1024x1024
npx gltf-transform resize input.glb output.glb --width 1024 --height 1024

# Resize only textures larger than 2048px
npx gltf-transform resize input.glb output.glb --width 2048 --height 2048

---

Step 4: Mesh Compression

Draco Compression

# Default Draco compression
npx gltf-transform draco input.glb output.glb

# Draco with quality settings
npx gltf-transform draco input.glb output.glb \
  --quantize-position 14 \
  --quantize-normal 10 \
  --quantize-texcoord 12 \
  --quantize-color 8

Draco Quantization Bits

Attribute	Default	High Quality	Aggressive
Position	14 bits	16 bits	11 bits
Normal	10 bits	12 bits	8 bits
TexCoord	12 bits	14 bits	10 bits
Color	8 bits	10 bits	6 bits

Higher bits = better quality, larger file. 14-bit position is sufficient for most models.

Quantize Without Draco

# Quantize attributes (reduces file size without Draco dependency)
npx gltf-transform quantize input.glb output.glb

Loading Draco-Compressed Models in Three.js

import { GLTFLoader } from 'three/addons/loaders/GLTFLoader.js';
import { DRACOLoader } from 'three/addons/loaders/DRACOLoader.js';

const dracoLoader = new DRACOLoader();
// ALWAYS set the decoder path — Draco uses WASM
dracoLoader.setDecoderPath('https://www.gstatic.com/draco/versioned/decoders/1.5.6/');
dracoLoader.preload();

const gltfLoader = new GLTFLoader();
gltfLoader.setDRACOLoader(dracoLoader);

gltfLoader.load('model-draco.glb', (gltf) => {
  scene.add(gltf.scene);
});

// ALWAYS dispose when done
dracoLoader.dispose();

Loading KTX2-Compressed Textures

import { KTX2Loader } from 'three/addons/loaders/KTX2Loader.js';

const ktx2Loader = new KTX2Loader();
// ALWAYS set the transcoder path — KTX2 uses WASM basis_transcoder
ktx2Loader.setTranscoderPath('https://cdn.jsdelivr.net/npm/three@0.160.0/examples/jsm/libs/basis/');
ktx2Loader.detectSupport(renderer);

const gltfLoader = new GLTFLoader();
gltfLoader.setKTX2Loader(ktx2Loader);

---

Step 5: Validation

Validation Checklist

After optimization, ALWAYS verify:

1. File size -- meets target budget from Step 1
2. Visual quality -- load in Three.js viewer, compare to original
3. Bounding box -- same dimensions as original (no scale errors)
4. Materials -- all textures load, no missing maps
5. Animations -- all clips play correctly (if applicable)
6. Performance -- measure draw calls, frame time

# Compare file sizes
ls -la original.glb optimized.glb

# Validate GLTF structure
npx gltf-transform inspect optimized.glb

# Check for errors
npx gltf-transform validate optimized.glb

---

LOD Generation

LOD Chain Strategy

LOD Level	Screen Coverage	Triangle Ratio	Distance
LOD0	>30% screen	1.0 (full)	Near
LOD1	10-30% screen	0.5	Medium
LOD2	3-10% screen	0.25	Far
LOD3	<3% screen	0.10	Very far

Generating LODs with gltf-transform

# Generate LOD chain
npx gltf-transform simplify model.glb lod0.glb --ratio 1.0
npx gltf-transform simplify model.glb lod1.glb --ratio 0.5 --error 0.001
npx gltf-transform simplify model.glb lod2.glb --ratio 0.25 --error 0.005
npx gltf-transform simplify model.glb lod3.glb --ratio 0.10 --error 0.01

Three.js LOD Implementation

import * as THREE from 'three';

const lod = new THREE.LOD();

// ALWAYS add LODs from highest to lowest detail
lod.addLevel(meshLOD0, 0);    // distance 0 = closest
lod.addLevel(meshLOD1, 10);   // switch at 10 units
lod.addLevel(meshLOD2, 30);   // switch at 30 units
lod.addLevel(meshLOD3, 80);   // switch at 80 units

scene.add(lod);

// In render loop — ALWAYS call update for LOD switching
lod.update(camera);

---

gltfpack Alternative

For quick one-shot optimization without a pipeline:

# Install
npm install -g gltfpack

# Optimize with meshopt compression
gltfpack -i input.glb -o output.glb -cc -tc

# Flags:
# -cc  mesh compression (meshopt)
# -tc  texture compression (KTX2/BasisU)
# -si N  simplify to ratio (0.0-1.0)
# -tp  texture power-of-two resize

NEVER mix gltfpack meshopt output with gltf-transform Draco -- the compression formats are incompatible.

Loading Meshopt-Compressed Models

import { MeshoptDecoder } from 'three/addons/libs/meshopt_decoder.module.js';

const gltfLoader = new GLTFLoader();
gltfLoader.setMeshoptDecoder(MeshoptDecoder);

---

Blender Export Optimization

When exporting from Blender to glTF:

Setting	Recommended Value	Why
Format	glTF Binary (.glb)	Single file, smaller
Apply Modifiers	ON	Bakes procedural geometry
Compression	OFF	Compress with gltf-transform after
Textures	Include	Embedded in .glb
Limit to	Selected Objects	Avoid exporting hidden/unused
+Y Up	ON	Three.js convention

NEVER enable Draco in Blender's exporter -- Blender's Draco implementation is outdated. ALWAYS compress with gltf-transform or gltfpack after export.

---

Node.js Scripting with gltf-transform API

import { NodeIO } from '@gltf-transform/core';
import { dedup, flatten, join, weld, simplify,
         textureCompress, draco, prune, quantize } from '@gltf-transform/functions';
import draco3d from 'draco3dgltf';
import sharp from 'sharp';

const io = new NodeIO()
  .registerExtensions(KHRONOS_EXTENSIONS)
  .registerDependencies({
    'draco3d.decoder': await draco3d.createDecoderModule(),
    'draco3d.encoder': await draco3d.createEncoderModule(),
  });

const document = await io.read('input.glb');

// ALWAYS run transforms in dependency order
await document.transform(
  dedup(),
  flatten(),
  join(),
  weld({ tolerance: 0.0001 }),
  simplify({ ratio: 0.5, error: 0.001 }),
  textureCompress({ targetFormat: 'ktx2' }),
  draco(),
  prune(),
  quantize(),
);

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

---

Reference Links

• references/methods.md -- gltf-transform API and Three.js loader methods
• references/examples.md -- Complete optimization pipeline examples
• references/anti-patterns.md -- Common optimization mistakes

Official Sources

• https://gltf-transform.dev/
• https://threejs.org/docs/#examples/en/loaders/GLTFLoader
• https://threejs.org/docs/#examples/en/loaders/DRACOLoader
• https://threejs.org/docs/#examples/en/loaders/KTX2Loader
• https://threejs.org/docs/#api/en/objects/LOD
• https://github.com/zeux/meshoptimizer