KitForma field notes

Prepare a GLB for web delivery

Make explicit changes and carry a report with the final asset.

Inspect the asset before reducing it

Start with a self-contained glTF 2.0 binary file. Embedded buffers and images make its dependencies explicit and keep local inspection independent of remote hosts. This workbench refuses external resources and unknown extensions instead of fetching unexpected files or discarding data that it cannot preserve. The source file remains unchanged.

Read the original inventory before choosing delivery settings. Mesh triangle counts describe mesh definitions and are not a rendered draw-call or instanced scene total. Materials, embedded images, skins and animation clips each contribute to the delivery cost. Unused resources can be an easy reduction, but a dense or already optimized asset may require a different authoring workflow.

Choose changes that match the destination

Default cleanup removes unreachable resources and can merge equivalent meshes, materials and textures with matching names. Empty scene leaf nodes, vertex attributes, single-color images and properties carrying extras are retained. These choices favor inspectability over aggressive deletion and do not simplify or quantize the geometry.

Texture resizing is a separate lossy choice. Animation removal is another explicit choice and does not bake a chosen animation frame. Keep both disabled when the destination needs the original visual detail and motion. A smaller file is useful only if it still communicates the intended model and behavior.

Validate and compare the delivery copy

The generated GLB is checked with the Khronos validator before a download is exposed. The audit includes source and output findings, sizes, material/image/animation inventories and applied changes. A structural validation pass is useful evidence but is not a guarantee of visual parity in all engines.

Preview both versions and then load the delivery copy in the actual destination renderer. Check every needed animation, material and scene, including transparency and normal maps. Retain the original as an authoring source and store the audit with the delivery copy. If the result grows, the report states that directly rather than presenting a fictional compression percentage.

The evidence behind the check

Sources & assumptions

Formats, algorithms and supported features are documented below. Check the stated scope and inspect your output before using it.

Reviewed 2026-10-19. Read our methodology or report an issue.

Essential local storage remembers this choice for up to 180 days. Your choice applies to this subdomain. Global Privacy Control and Do Not Track override optional measurement.

Advertising has a separate choice. Use Advertising preferences in the footer to allow or reject the optional guide-page banner.