Mesh decimation
Polygon reduction with controllable intensity – shrink dramatically while preserving silhouette and form.
3D ToolingWebGLDeveloper Tool
ShrinkModel
Compress heavy 3D models, then walk through the result side by side with the original.
- Role
- Engineering & 3D pipeline design
- Industry
- 3D & creative tooling
- Timeline
- 10 weeks
- Scope
- Compression · Comparison · Viewer
- Platform
- Web – Pipeline · Viewer
- Status
- Internal tool
Overview
The story in brief
ShrinkModel is a web application for compressing 3D models – and for proving the compression held up. Upload a heavy model, tune how aggressively meshes and textures are optimized, then open the original and the compressed result in two viewers side by side and judge the difference for yourself.
Both frames support first-person mode, so you can walk through the compressed model exactly as you would the original. For environment and property creators, that walk is the real quality check: detail that survives a screenshot often falls apart at eye level.
It works just as well as a viewer alone. Plenty of sessions never compress anything – they open a model in the browser and walk it.
02
The challenge
Production 3D assets routinely run to hundreds of megabytes, and compressing them is a leap of faith: the size drops, but nobody can say what it cost until the model is back in a scene.
Naive compression destroys exactly what creators need – silhouettes, surface detail, texture fidelity. A single-frame preview hides that, because there's nothing to hold it against.
Judging a 3D environment from the outside misses the point. Creators experience their work from inside it, and quality problems only show up at walking height and walking distance.
03
The approach
We built compression as a tunable pipeline: mesh decimation and texture optimization run with adjustable presets, so the operator chooses the trade-off between size and fidelity per asset – and sees the result before committing.
Comparison is the product's other half. Original and compressed load into two synchronized frames, so the difference is visible directly rather than remembered between two separate loads.
First-person navigation puts creators inside the model. Environment and property builders walk the compressed scene at human scale, which is the only way to catch what a static preview politely hides.
Every processed asset keeps its lineage – source, settings, and output size – so teams can reproduce a result or step back to the original at any time.
04
What we built
The pipeline, the two frames, and the walkthrough that proves the compression held.
01/08
compress
Mesh decimation
Texture optimization
Textures are resized and re-encoded to GPU-friendly formats, often the biggest single saving in the file.
Quality presets
Preview, review, and archive presets encode sensible trade-offs – with per-asset overrides when a model needs care.
In-browser viewer
Models open directly in the browser – no plugins, no desktop software. Use it to compress, or just to view.
First-person walkthrough
Walk the model at human scale instead of orbiting it – how environment and property creators actually judge their work.
Side-by-side comparison
Original and compressed open in two frames at once, so what the optimization cost is visible rather than remembered.
Batch processing
Queue whole asset sets through the same settings – consistent results across a project without per-file babysitting.
Shareable links
Every processed model gets a link that opens straight into the viewer – reviews start in one click.
Mesh decimation
Polygon reduction with controllable intensity – shrink dramatically while preserving silhouette and form.
Texture optimization
Textures are resized and re-encoded to GPU-friendly formats, often the biggest single saving in the file.
Quality presets
Preview, review, and archive presets encode sensible trade-offs – with per-asset overrides when a model needs care.
In-browser viewer
Models open directly in the browser – no plugins, no desktop software. Use it to compress, or just to view.
First-person walkthrough
Walk the model at human scale instead of orbiting it – how environment and property creators actually judge their work.
Side-by-side comparison
Original and compressed open in two frames at once, so what the optimization cost is visible rather than remembered.
Batch processing
Queue whole asset sets through the same settings – consistent results across a project without per-file babysitting.
Shareable links
Every processed model gets a link that opens straight into the viewer – reviews start in one click.
01 / 08
05
Under the hood
Hover a stack group to trace its path in the system
- Viewer
- Three.jsReactWebGL
- Upload console
- ReactVite
- Pipeline API
- Node.jsglTF toolchain
- Processing
- DracoKTX2 / Basis
- Infrastructure
- DockerObject storageCI/CD
Architecture
- conversion_pathCompression runs as queued pipeline jobs, so heavy processing never blocks uploads and batches scale horizontally.
- conversion_pathThe viewer streams compressed glTF with Draco and KTX2 payloads, keeping first render fast even on modest connections.
- conversion_pathEvery output stores its source and settings lineage, making results reproducible and originals always recoverable.
06
The outcome
Internal tool · compression + viewer
75%
Typical size reduction
10 wks
Build timeline
2 frames
Original · Compressed, side by side
0
Installs needed to view
Compression stopped being a leap of faith: original and compressed sit side by side, so the cost of every setting is visible before anything ships.
Environment and property creators walk the compressed model in first person and judge it the way their audience will experience it.
Heavy 3D assets open in a browser – to compress, or simply to view and explore, with originals always one step away.
Next