A-Frame Low Poly Models

A-Frame loads GLB models through the `gltf-model` component, usually preloaded inside `<a-assets>` so the scene can manage loading before display. GLB is preferable to loose glTF because a WebXR page should avoid a pile of separate JSON, buffer and texture requests. Export in meters and keep pivots meaningful; a model that appears correctly scaled in Blender can still feel enormous in VR if its real-world dimensions were never checked.

WebXR budgets are stricter than ordinary desktop web pages because the scene may render stereoscopically at high refresh rates. Repeated props should target 100 to 600 triangles, interactive props around 1k to 3k, and characters around 3k to 10k for standalone headset comfort. Texture atlases should stay lean, commonly 512px for small props and 1024px for hero objects, because mobile headset memory and thermal limits matter.

Scene structure matters in A-Frame because every entity carries component overhead. Use one GLB for a combined static kit when objects never move separately, and separate entities only when interaction, animation or collision requires it. For collisions and interactions, prefer simple boxes, spheres or low-detail proxy meshes rather than the visible mesh. A faceted decorative model may look cheap, but triangle-accurate ray or physics tests still cost real CPU time.

Test on the actual target mode: inline browser, mobile AR, desktop VR or standalone headset. A model that feels fine on a laptop can miss frame rate in headset because it renders per eye and competes with tracking overhead. Compress GLB carefully, avoid oversized transparent materials, and keep material count low. A-Frame is forgiving for prototypes, but production WebXR assets need the same discipline as any real-time engine.