Where this sits
A visionOS piece is built in Xcode and Reality Composer Pro. Nothing replaces that, and this page is not going to pretend otherwise. If somebody tells you their generative tool exports a shippable visionOS app, ask to see the project file.
What happens before that is concept, scene design and previs — deciding what the space is, how it is lit, what the viewer sees when they turn around, and whether the idea is worth the six weeks of engineering that follow. That is the part this product is for, and it is the part where most projects are won or lost.
A working shape
Design the space here
Build the scene as a graph: one place, consistent light, several views that agree with each other. This is where the look is decided and where changing it is still cheap.
Board the experience
Use the director stage to lay out the beats a viewer moves through, with duration and notes on each. This becomes the document the engineers actually work from.
Export frames as reference
Take the approved frames out as a reference pack for whoever models and builds the scene properly.
Build in the Apple tools
Reality Composer Pro and Xcode do what only they can do. The handoff is a reference pack, not a project file, and treating it as anything else will disappoint everyone.
Come back for revisions
A change of direction is cheaper to test here than to rebuild there. This is the main reason to keep the graph alive after handoff rather than treating it as a phase you finished.
What this does not export
If that changes it will appear on the status page when it works, not before. We would rather this page be useful and accurate than optimistic — you are making resourcing decisions from it.
Questions from teams shipping on visionOS
Can I generate a skybox or an environment map?
You can generate imagery that a technical artist can turn into one, and people do. There is no direct equirectangular or cubemap export, so expect a manual step.
Is it useful for a mostly-2D visionOS app?
Yes, and arguably more so. Window and panel layouts, iconography and marketing material are ordinary design work, and the consistency tooling applies directly.
What about stereo pairs?
Not supported. Generating a convincing left and right eye that agree is exactly the depth-consistency problem the roadmap calls out, and a naive offset produces something that gives people headaches.
How do I keep a look consistent between here and Reality Composer?
Treat the graph as the reference of record — palette, light direction and material vocabulary as nodes — and match to it rather than to individual frames. Frames drift; rules do not.