GL Transmission Format

STL Stores Triangles. glTF Stores an Entire 3D World. What's the Difference?

A glTF file describes a 3D scene the way a browser builds a web page: a top-level container that holds a hierarchy of nodes, each node carrying a transform, a mesh reference, or a camera/light. Meshes reference primitives; primitives reference accessors; accessors describe typed views into raw binary buffers that map one-to-one with how a GPU expects vertex data.

That hierarchy is the design. Khronos chose a JSON manifest so that a browser or game engine can parse the scene graph and start issuing GPU uploads before the heavy geometry has finished downloading. By the time the .bin buffer arrives, the renderer already knows which accessor spans which bytes, what component type (float, uint16, uint32) they represent, and which GPU attribute slot they feed. No parsing of the binary is needed — just allocate a buffer on the GPU and memcpy.

This is what makes glTF the 'JPEG of 3D.' JPEG decouples metadata from image data; glTF decouples scene structure from geometry. STL was built for a 3D printer — one mesh, no metadata. glTF was built for a browser's rendering pipeline — a world with cameras, lights, skins, animations, and materials, streamed in parallel. That choice shapes how the file is packaged — and glTF gives you two ways to package it.

One Scene, Two File Formats. When Do You Use .gltf and When .glb?

glTF ships in two flavors. The .gltf file is a JSON manifest that references separate .bin files for binary buffers and .png/.jpg files for textures. The .glb file packages everything into a single binary container.

A GLB file begins with 12 magic bytes: 67 6C 54 46 (ASCII glTF), followed by a 4-byte version (currently 2), followed by a 4-byte total file length. After the header come two chunks. Chunk 0 holds the JSON manifest — marked with chunk type 0x4E4F534A (JSON in ASCII). Chunk 1 holds the binary buffer — marked with chunk type 0x004E4942 (BIN\0 in ASCII). Each chunk begins with 4-byte length + 4-byte type before the payload. The JSON chunk is padded with ASCII spaces to a 4-byte boundary; the BIN chunk is padded with null bytes.

The design trade-off is explicit. .gltf plus sidecar files works well for authoring and version control — a Blender artist can commit the JSON, update individual textures, and diff changes meaningfully. .glb works better for distribution — one file, no missing-sidecar problems, trivial drag-and-drop for AR previewers. Most 3D pipelines produce .glb for shipping and .gltf for source. The data inside is identical; the packaging is the choice. The packaging is the choice. The materials rendered inside are where glTF caught up to modern expectations — without ever changing its version number.

The Core Format Hasn't Changed Since 2017. Why Does It Still Render Modern Materials?

A glTF material isn't a color. It's a physically-based rendering recipe with 40+ Khronos extensions to refine it.

The core material model is metallic-roughness PBR, specified in glTF 2.0: a base color (RGB), a metallic factor (0-1), a roughness factor (0-1), a normal map for surface detail, an ambient-occlusion map for contact shading, and an emissive color for self-illumination. That base covers about 80% of real-world materials — wood, plastic, painted metal, fabric.

The remaining 20% ships as KHR_* extensions. KHR_materials_unlit disables lighting for icons and UI elements. KHR_materials_clearcoat adds a second specular layer for car paint or nail polish. KHR_materials_transmission handles glass and clear liquids. KHR_materials_volume models absorption through translucent materials. KHR_materials_sheen simulates velvet and brushed fabric. KHR_materials_iridescence reproduces soap-bubble and beetle-shell color shifts. KHR_materials_anisotropy captures brushed metal.

Compression extensions attack a different problem. KHR_draco_mesh_compression reduces geometry size by 90% or more through vertex prediction. KHR_mesh_quantization lets attributes use lower-precision integer types for similar gains with no dependency on external decoders. KHR_texture_transform adds UV offset and tiling without modifying the source texture.

Khronos froze the glTF 2.0 core spec in 2017. Everything since — seven years of real-time rendering innovation — has shipped as extensions. The version never changed because it didn't need to. Modern renderers speak glTF 2.0 plus whichever extensions the viewer supports, and the rest degrade gracefully to the metallic-roughness base. The format stayed still. The ecosystem moved. And nowhere did that movement go further than mobile AR.

Point Your Phone at a QR Code. A 3D Model Floats in Your Kitchen. How?

Point your phone at a QR code and a 3D model floats in your kitchen. It's glTF — until iOS converts it to USDZ.

Google built the cross-platform AR stack around glTF. Android's Scene Viewer accepts GLB URLs directly from web pages; Google's <model-viewer> web component renders interactive 3D previews with a single HTML tag and delegates to Scene Viewer for AR mode. Three.js, Babylon.js, and PlayCanvas all ship native glTF loaders as first-class features. Unreal Engine, Unity, Godot, Blender, Sketchfab, and Adobe Dimension export to glTF for web delivery.

Apple's Quick Look is the exception — iOS renders USDZ (Pixar's Universal Scene Description in a ZIP), not glTF. But even Apple's pipeline accepts glTF upstream. Tools like Reality Composer Pro and adopted web components convert glTF to USDZ automatically when targeting iOS AR.

Museums and cultural institutions have turned glTF into a heritage-preservation format. The Smithsonian's 3D Digitization Program publishes artifact scans as glTF for web virtual tours. The British Museum's CyArk partnership captures UNESCO World Heritage sites as glTF assets. Google Arts & Culture hosts thousands of 3D-scanned artifacts worldwide. E-commerce drove mainstream adoption: IKEA's Place app, Amazon's 3D product previews, and Shopify's AR Quick Look integration all use glTF/GLB assets to render furniture and products in a customer's own room before purchase.

glTF quietly became the cross-platform 3D standard the web always wanted.