What 3D formats does Zlim support?

Zlim converts glTF/GLB, OBJ, STL, PLY, FBX, Collada (DAE), USD/USDZ, 3MF, STEP, Creo, and CATIA to optimized GLB. IGES, SolidWorks, DXF, 3DS, and VRML/X3D are coming soon.

How much smaller will my model be after optimization?

Zlim typically achieves 70–80% file size reduction through mesh quantization, Draco compression, and texture optimization — while preserving visual fidelity.

What does Zlim optimize besides polygon count?

Zlim optimizes the whole budget, not just polygons: faces (mesh decimation and quantization), draw calls (merging meshes and atlasing materials), and texture memory (KTX2/Basis transcoding). Each is measured and reported per job.

Three integration paths: (1) local CLI — run 'zlim optimize model.fbx -o model.glb -p balanced' (no API key); (2) local MCP server — connect AI agents like Claude; (3) hosted REST API — create a job, upload the file, and poll until it succeeds.

// format optimizer

Any 3D format in,
optimized model out.
One API.

Stop juggling Blender, Pixyz, gltfpack, and a zoo of plugins. Zlim ingests the dirty formats — STEP, CATIA, FBX, USD, and more — and returns one clean, optimized GLB. Deterministic. Measurable. CLI, MCP, or API.

70–80%

file size reduction

input formats

clean output (GLB)

Get an API key Read the docs

The 3D format tax

Format chaos

Every tool exports its own thing — STEP from CAD, FBX from your DCC, USD from the pipeline. Nothing speaks to everything.

Bloat

Raw exports ship 5–10× heavier than they need to be: uncompressed geometry, full-res textures, thousands of stray draw calls.

Broken files

Dirty CAD, FBX, and USD exports arrive with missing materials, inverted normals, or geometry no viewer will open.

Won’t load on web or AR

Authoring formats aren’t real-time formats. Browsers, AR, and game engines need GLB — and they need it small.

Why Zlim

Dirty-format breadth is the moat

Only Zlim ingests the formats nobody else tames — STEP, CATIA, Creo, FBX, USD, 3MF — and returns one clean, optimized GLB in a single pass. No Blender round-trips, no plugin zoo.

Deterministic quality

70–80% smaller. Not approximately smaller. Measured per job via mesh quantization, Draco compression, and texture mipmap generation.

API · CLI · MCP server

Integrate into any workflow: REST API for backends, CLI for pipelines, and an MCP server so AI agents can invoke 3D optimization directly.

Starts free, scales with you

Start free with 25 optimizations a month. Upgrade to Pro at $29/mo for more volume, every format, and larger files. Simple plans, no per-seat pricing.

What Zlim optimizes — the whole budget

−70–80%

Faces

Tolerance-bounded decimation and vertex quantization remove redundant geometry without changing the silhouette.

Fewer

Draw calls

Mesh merging and material atlasing cut the number of GPU binds — the lever that actually moves frame rate on the web and in AR.

KTX2

Texture memory

Textures transcoded to KTX2/Basis stay GPU-compressed in VRAM — the hidden half of the budget that crashes mobile and AR.

Poly count is one lever of three. Zlim optimizes all of them — and reports each, measured, on every job.

Before / after — one measured job

A representative product model on the default profile. Every number is reported per job — deterministic and reproducible.

Metric	Before	After	Reduction
Faces	1,240,000	286,000	−77%
Draw calls	340	12	−96%
Texture memory	96 MB	18 MB	−81%
File size	45 MB	9.2 MB	−80%

Same look, every lever down. Results depend on the source model; the reduction is always measured and returned with the job.

How it works

Upload your model

Create a job on the hosted API and upload your file, or point the local CLI at a file. Any supported format.

bash

curl -X POST https://zlim.ai/api/v1/jobs \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "filename": "model.fbx", "options": { "preset": "balanced" } }'

Zlim optimizes

The engine runs mesh decimation, quantization, Draco compression, and texture optimization. Typically completes in seconds.

Download the GLB

Poll the job status, then fetch the result URL. A signed download link returns the optimized GLB — 70–80% smaller.