Build lighter mesh variants for real-time projects.

Limited-time launch offer: use code smartstart50 at checkout for 50% off the 12 USD base price (6 USD before applicable taxes). Ends October 29, 2026 at 11:59 PM CDT. Applies only to Smart LOD Generator.

Smart LOD Generator creates progressively reduced copies of selected static mesh assets in Blender. Set the number of levels and a triangle ratio, review UV and feature constraints, then generate an organized LOD chain.

For artists preparing assets for

• Games: static environment props, equipment and background assets.
• Real-time viewers: product and industrial mesh variants for different detail budgets.
• Architectural visualization: lighter static props for larger scenes.

Five stages, one chain

Small-island removal, coincident-vertex welding, planar region merging, coordinated edge-loop reduction and UV-aware QEM collapse work toward each level's requested triangle budget. Enable only the stages that suit your asset. Small-island removal is based on size and can remove visible details; it does not test occlusion.

Control what matters

Evaluate all UV layers or only the active one. Tune local geometry/UV tolerances, preserve feature constraints, protect important vertices with a vertex group and inspect candidates with the analysis heatmap. Individual tools offer focused cleanup on selected mesh data.

Organized, reviewable output

Generate independent LOD mesh copies in an owned collection. Keep the source geometry, optionally arrange copies side by side and inspect triangle counts in the report. Regeneration replaces only tagged outputs belonging to the source; name conflicts with unowned objects are blocked.

New in 1.1

• Custom Triangle Budgets: enter a requested triangle count for each LOD level instead of using one ratio. Targets are validated before replacing the chain.
• LOD Inspector + CSV: compare current base-mesh counts with saved generation targets and export a local report. The report includes project names; review it before sharing.

1.1 also bounds temporary candidate-processing allocations, reuses face lookup keys and restores heatmap color/viewport settings. The included isolated CPU microbenchmark uses less traced temporary memory with a small timing increase; no universal speedup is promised.

Measured synthetic demo

A subdivided beveled-box fixture with two UV maps went from 9,408 triangles to 1,176 at LOD3: 8,232 removed (87.5%). LOD1: 4,666; LOD2: 2,326. Tested with 1.0 on Blender 5.2.1 CPU, ratio 0.5, Strict UV on and Relax for Target off. This controlled example is not a representative production benchmark or a guaranteed reduction/visual-error result. Full settings are in Documentation and the included measurement JSON.

Know the limits

Use static local mesh assets. Shape keys, armatures, linked/overridden data and multi-scene sources are unsupported. In-place tools and heatmap analysis require single-user mesh data with visible modifiers baked. Targets can be missed or slightly overshot. UV checks are local estimates, not a guarantee of identical textures, shading or a global error bound; approximation can accumulate across levels. Review silhouettes, textures and exports.

This prepares geometry; it does not add game-engine distance switching, export automation, rigged-character LODs or retopology. Generating all levels adds geometry to the scene; it does not by itself reduce project memory, file size or render time.

Compatibility and included files

Blender 5.0 minimum. CPU workflow validated on Windows with Blender 5.0.0 and 5.2.1 LTS. Windows x64 Blender Extension ZIP with Python source and GPL license, detailed English guide, FAQ, changelog, CSV report example, measured demo data and validation reports. Version history retains the 1.0 release. No external dependencies are required for CPU use. Optional unbundled CuPy/CUDA cost acceleration is unvalidated; no fixed speed or reduction percentage is promised.

Main crate graphics illustrate concepts and are not included production assets or software-interface captures. The separate checker-box comparison uses actual synthetic-fixture renders with measured counts.

License: GPL-3.0-or-later for Python code. Your own processed models keep their existing rights.