The checklist below will help you avoid making multiple revisions when delivering models for AR Quicklooks on the Variant platform. Please review it carefully when starting to model, and before uploading
And for each standard material, or material variant:
In exporters this is sometimes referred to as ‘GLTF Separates’. The important thing is that it is not a GLB, or an embedded GLTF (where textures are all packed in a single .gltf file).
Valid GLTFs don’t use spaces or special characters (e.g. []\|}{[email protected]#$%^&*()=`) in their filenames, so the converter will fail if they are used. Rename textures and your project files to avoid them. E.g. ‘wood [red!].png’ should become ‘wood_red.png’.
An ideal size for a single Variant that balances quality and download speed is around 8MB. If you have 3 different color variants, only one will be downloaded by the user at a time, so the files you upload can be larger than 15MB, as long as the user will download less for a single model.
Use a workflow based on metres as a unit if possible. Check the model scale on the Variant site, or via Sketchfab AR.
This is vital to avoid the model hovering above the ground, or misaligned rotation. The origin point will be placed on the floor where the user taps, so make sure its the lowest point of the model, and at the center point.
Small items like screws, plates, etching etc. should use textures instead of polygons.
These files will be downloaded over the web, and users expect their model to load roughly as fast as a web-page. 70-85% quality in photo editing packages is normally the ‘sweet spot’.
If you are supplying only diffuse and normal maps, you are likely not using a PBR workflow, which should include details in the metal and roughness channels.
The iOS and Android renderers assume the colour texture is ‘albedo’ which is the pure color of the surface, with no shadow or shading added (these are added when AO, normals and lighting are applied). If you are not using an albedo-based approach, colours will seem dull, desaturated and too dark in AR, as shadows are being applied twice.
To keep file-sizes down, and to ensure the compressed file data isn’t being spread over too many pixels (resulting in JPG artifacts), non diffuse files should be lower resolution. E.g. it is better to have fewer, more accurate pixels in a normal texture, than big JPG artifacts that change the appearance of the surface.
USDZs cannot use alpha modes other than Opaque or Blend, and all Variant models must be convertible to USDZ.