Skip to content
The Technical ArtistArt. Code. Pipelines. Games.
section 09
simplygon
optimisation

Automate Asset Reduction And Review The Result

Workingguide~5 min read

Simplygon can automate LOD generation, proxy meshes, material baking, and impostors. These pages focus on using it from Python and Maya, choosing settings for the problem, and checking the output. Start with a representative asset before committing to a larger batch.

A wireframe lantern becomes three progressively simpler LOD versions beside a texture atlas.
simplygon - mesh reduction and texture atlases
On this page What It Is Why Use It Method Map Licence Path Workflow Shape

01What Simplygon Is

Simplygon is a mesh optimisation SDK. It provides several ways to reduce rendering costs, including mesh reduction, remeshing, and material merging. The useful method depends on the asset and the bottleneck, so compare the output visually and measure it in the target engine.

You can drive it from C++, C#, or Python, and there are plugins for Maya, Max, Unreal, and Unity. Everything on these pages uses the Python API from SDK 10.4, because Python is where TAs live.

A processing job takes a scene and a set of settings, then writes an output scene. Keep the source, settings, and SDK version together so you can reproduce the job and compare results after a change.

mental model

Consider Simplygon when asset optimisation needs to become a repeatable pipeline step. Test whether the results meet your quality and performance targets before adding it to a batch or build process.

02Why Use It

Repeatability is a useful part of the workflow. When a source mesh changes, a scripted process can regenerate its outputs from the same settings. You’ll still need to review those outputs, particularly if the new source differs substantially.

  • Scale: point it at a folder, come back to a folder of LODs. Review the batch report and inspect a sample of the outputs.
  • Frame time: triangles, draw calls, bone counts, hidden interior geometry - choose the process that addresses the measured cost.
  • Iteration: settings live in JSON files you version-control. Budgets change? Edit the file, rerun, done.
  • Consistency: every asset gets the same rules, so LOD2 never depends on who made it or what day it was.
  • The weird stuff: dense scans, modular characters, foliage billboards, HLOD proxies - there's a dedicated workflow for each, so you're not bending the wrong tool.

03The Optimisation Methods

The right method depends on the cost you're trying to remove. Don't pick "reduction" by reflex. Pick the process that attacks the bottleneck.

Method Use It For Preserves Tradeoff
Triangle reduction General static meshes and far LODs where topology doesn't need to stay artist-friendly. Overall silhouette and existing materials, depending on settings. Can create long triangles or topology that's unpleasant for editing.
Quad reduction Subdivision-like or deformation-sensitive meshes where quad structure matters. More editable topology flow than triangle reduction. Less aggressive than triangle reduction and not right for every mesh.
High-density reduction Very dense scans or CAD-like assets that need large triangle-count drops. High-level shape while simplifying extreme input density. Needs careful validation of normals, UVs, and baked maps.
Remeshing Proxy meshes, HLOD-style assets, scans, and assets with messy topology. Visual surface through new topology and baked material channels. Replaces original topology, so it's not for deformation-friendly outputs.
Aggregation Merging many objects to reduce draw calls while keeping their triangle shapes. Original mesh topology more than remeshing does. Triangle count may not drop unless paired with culling or reduction.
Material merging Atlas creation, fewer material slots, lower draw-call pressure, and proxy/HLOD baking. Rendered look through colour, normal, roughness, metallic, opacity, or custom casters. Needs channel discipline, UV validation, and texture-size budgeting.
Visibility culling Removing occluded or never-seen interior/backside geometry. Visible surfaces from configured cameras or visibility volumes. Bad camera/occluder setup can remove geometry players later see.
Impostors and billboards Very distant LODs for vegetation, props, architecture, and fixed-view assets. Appearance from baked views rather than real geometry. Works poorly when the player can inspect the asset closely or from unusual angles.
Bone reduction Skeletal meshes where far LODs need fewer bones or lower skinning cost. Important deformation behaviour, depending on bone settings and weights. Must be checked in animation, not only in bind pose.
Clustered meshlets Renderer-specific mesh clustering or modern runtime rendering experiments. Mesh content packaged for cluster-oriented rendering. Needs engine-side support and a clear rendering reason.

04Getting A Licence

Short version: everything starts at simplygon.com - the download, the licence, the docs. Licensing terms shift over time, so treat their site as the source of truth, not this page (or any page that isn't theirs).

As of mid-2026 they offer a per-game-title licence, an evaluation path, and discounted pricing for smaller indie studios - check the current terms before planning a budget.

  1. Head to simplygon.com and read the pricing and evaluation information.
  2. Requesting an eval? Use your company email and be specific about what you're building - "we need LODs for a 2,000-asset open world" gives the vendor useful context for the request.
  3. Once you're in, grab the SDK installer - it bundles the Python bindings and the DCC plugins, and detects your Maya/Max versions on its own.
  4. Before you build anything on top of it, make sure the SDK actually initialises from Python on your machines. Confirm this before writing tools that depend on the installation.
production warning

Don't build half your pipeline on a trial install and then ask whether the licence covers your build farm. Check the terms first - redistribution, CI machines, seat counts. Include build machines and distribution needs in that check.

05The Site Workflow

These pages follow the order most TAs actually learn this stuff in:

  • Standalone Usage - pure Python, no DCC open. Import a scene, process it, inspect the output, then try a folder of assets.
  • Maya Usage - the artist-facing side. Load the plugin, grab the selection, process it, drop the result back in the scene without anyone leaving the viewport.
  • Tips - checks to make before using the output: picking the right method, targets that don't lie, keeping materials intact, and why you always test characters in motion.