OMNIFORM SYSTEMS Benchmark assets
OMNIFORM SYSTEMS · 3D ASSET PREPARATION INFRASTRUCTURE

Raw 3D data
is not simulation data.Omniform closes the gap.

Omniform turns CAD, scans, marketplace models, and heavy mesh assets into simulation-ready outputs for robotics, synthetic data, digital twins, and real-time engines.

Automated cleanup · Topology repair · Adaptive optimization · LODs · Collision proxies · Metadata · QA reports

Benchmark your assets Send us a messy model
§ 01 · PROBLEM CREATED FOR:
HUMAN VISUALIZATION
DESIGN REVIEW
DOCUMENTATION
RENDERING

NOT CREATED FOR:
PHYSICS SIMULATION
ROBOT TRAINING
SYNTHETIC DATA
DIGITAL TWINS
§ 01 · PROBLEMp. 02

The world has 3D data. It does not have simulation-ready 3D data.

Raw files are too dense. Scans are noisy. Marketplace assets are inconsistent. Photogrammetry creates broken topology. Meshes often arrive without LODs, collision geometry, clean hierarchy, usable scale, or simulation metadata.

A human can look at a messy 3D model and understand it. A simulator cannot. A physics engine cannot. A robot policy cannot. Before raw assets can enter Isaac Sim, Omniverse, MuJoCo, Unreal, or Unity, they need to be cleaned, simplified, repaired, structured, and validated.

Today, that preparation layer is still largely manual.

FAILURE MODES TOO DENSE
NOISY TOPOLOGY
BROKEN NORMALS
NO LODs
NO COLLISION GEO
WRONG SCALE
WRONG PIVOT
NO SEMANTIC STRUCT
MISSING METADATA
BROKEN MATERIALS
§ 02 · SOLUTION NOT A SERVICES FIRM.
NOT GENERIC MESH
COMPRESSION.

SIMULATION DATA
INFRASTRUCTURE.
§ 02 · WHAT OMNIFORM DOESp. 03

Omniform automates the preparation layer between raw spatial data and simulation-ready assets.

Our pipeline analyzes incoming 3D assets, repairs geometry, identifies simulation-relevant structure, reduces unnecessary complexity, generates LODs and collision proxies, preserves useful metadata, and returns engine-ready outputs with a QA report.

We are not building another visual compression tool. Omniform is built for machine use: physics, collision, affordances, metadata, and runtime performance. The output is measured, validated against the target environment, and reported before it leaves the pipeline.

POSITION THE MACHINE-
READINESS LAYER
FOR SPATIAL DATA

THE INFRASTRUCTURE
LAYER BETWEEN
RAW 3D DATA AND
SIMULATION-READY
ASSETS
§ 03 · HOW IT WORKSp. 04
INPUT01
Raw spatial data
CAD EXPORTS (STEP/IGES)
LIDAR CAPTURES
PHOTOGRAMMETRY
3D SCANS
FBX · OBJ · GLB · USD
MARKETPLACE MODELS
ENGINEERING MESHES
OMNIFORM PIPELINE02
Asset preparation
PREFLIGHT ANALYSIS
GEOMETRY CLEANUP
TOPOLOGY HEALING
SEMANTIC-AWARE OPTIM.
LOD GENERATION
COLLISION PROXIES
PBR REBAKING
HIERARCHY CLEANUP
METADATA PRESERVATION
QA REPORTING
OUTPUT03
Simulation-ready output
ISAAC SIM
NVIDIA OMNIVERSE
MUJOCO
UNREAL ENGINE
UNITY
USD · FBX · GLB
ENGINE-VALIDATED
§ 04 · APPROACH MOST TOOLS:
LOOKS GOOD
TO HUMAN EYE

OMNIFORM:
USABLE BY SIM
COLLIDABLE
TRAINABLE
PHYSICS-READY
§ 04 · BUILT FOR MACHINESp. 05

Built for machines, not just the human eye.

Most optimization tools focus on how a model looks. Omniform focuses on whether the model can be used by a simulator, physics engine, robot training pipeline, or digital twin system. Visual fidelity is a constraint Omniform preserves — it is not the objective.

PRESERVE
Matters to sim
CONTACT SURFACES
HANDLES · RIMS
AFFORDANCES
SILHOUETTES
ARTICULATION ZONES
MATERIAL BOUNDARIES
NAMED COMPONENTS
REDUCE
Over-modelled
FLAT SURFACES
DENSE SCAN NOISE
EXCESSIVE SUBDIVISIONS
HIGH-POLY CLUTTER
NON-CONTACT INTERNALS
REMOVE
Hidden / irrelevant
HIDDEN INTERNAL GEO
DUPLICATE SURFACES
BROKEN FRAGMENTS
IRRELEVANT MICRO-DETAIL
SIM-IRRELEVANT GEO
RESULT TRIANGLE COUNT:
REDUCED

FILE SIZE:
REDUCED

VISUAL FIDELITY:
PRESERVED

IMPORT STATUS:
CLEAN

ALL MEASURED.
ALL REPORTED.
§ 05 · PROOF EVERY PROCESSED
ASSET SHIPS WITH
A BEFORE / AFTER
REPORT.

NO BLACK BOX.
NO CLAIMS WITHOUT
NUMBERS.
§ 05 · BENCHMARK REPORTp. 06

Every asset should prove its own improvement.

Omniform outputs are measured, not guessed. Each processed asset can return with a before/after report showing what changed, what was preserved, and what still needs review.

BEFORE · SOURCE ASSETS
AFTER · OMNIFORM OUTPUT
ASSETS TESTED
Mixed source library
19 assets — CAD · MESH · SCAN · MARKETPLACE
TRIANGLE COUNT
29.1M triangles (total)
9.4M triangles — 67.6% reduction
DRAW CALLS
6,431 calls (total)
38 calls — 99.4% reduction
GEOMETRY WEIGHT
Source baseline
3.1× lighter (aggregate)
DRAW CALL EFFICIENCY
Source baseline
169× fewer (aggregate)
STRONGEST CASE — TRIANGLES
5,520,900 triangles
97,500 tri — 98.2% (4090_Gpu_V1)
STRONGEST CASE — DRAW CALLS
2,741 calls
2 calls — 99.9% (4090_Gpu_V1)
PHYSICAL LOD VALIDATION
Not yet measured
Next benchmark layer — contact · collision · articulation
WHAT SHIPS OPTIMISED MESH
LOD CHAIN
COLLISION PROXY
CLEANED HIERARCHY
NORM. SCALE + AXES
PRESERVED MATERIALS
METADATA SIDECAR
ENGINE-READY EXPORT
QA REPORT
§ 06 · INTERACTIVE BENCHMARKp. 07

Inspect the same asset before and after Omniform preparation. Switch views to compare geometry density, collision readiness, and LOD output.

Loading viewer
Interactive preview unavailable
WebGL is not available in this browser. Benchmark metrics are still shown below.
Asset preparation report
Asset typeMarketplace model
Target workflowIsaac Sim / Omniverse
Processing modeAutomated first pass
72%
Triangle reduction
58%
File size reduction
4
LOD levels generated
Before After Delta
Triangles2.4M672K−72%
File size840 MB353 MB−58%
Materials9632−67%
LODsNoneLOD0–LOD3Generated
CollisionMissingProxy generatedReview-ready
ImportCleanup requiredReview-readyImproved
What changed
  • Simplified simulation-irrelevant geometry
  • Preserved visible silhouettes and functional surfaces
  • Generated LOD chain (LOD0–LOD3)
  • Created collision proxy geometry
  • Preserved and rebaked materials
  • Returned QA notes for remaining review items
Performance depends on scene structure, materials, physics settings, and target hardware. Results are asset-specific.

Send us a representative asset and we'll show the before/after report.

§ 06 · BENCHMARK SEND 5–10
NON-CONFIDENTIAL
MESSY ASSETS.

CAD EXPORTS
SCANS
MARKETPLACE MODELS
HEAVY MESHES

WE RETURN:
PROCESSED ASSETS
+ FULL REPORT

NO COMMITMENT.
REPLY FROM A
FOUNDER.
§ 06 · ASSET BENCHMARKp. 07
Send us the kind of asset your team does not want to clean by hand.

Share 5–10 non-confidential messy assets — CAD exports, scans, marketplace models, or heavy meshes. We'll return optimized outputs with before/after metrics, import notes, LODs, collision proxies where applicable, and a simulation-readiness report.

Benchmark your assets Send us a messy model
BUILT BY A TEAM WITH EXPERIENCE IN COMPLEX SCIENTIFIC GEOMETRY · CERN-SCALE REAL-TIME 3D ENVIRONMENTS · GEOMETRY MODELING · ML ENGINEERING · STARTUP EXECUTION
CONTACT HELLO@
OMNIFORM.SYSTEMS

RESPONSE FROM
A FOUNDER
WITHIN 24H.