AI image generation: preserve dimensions and orientation when editing design-tool exports (e.g. Redx)

I build custom decks and use Redx to draw out 3D designs for clients. When I upload one of my Redx designs into Handoff and ask for a rendering, the AI doesn't preserve the dimensions or orientation of the deck I actually drew.

For example: I had a 20x10 walkout deck with stairs going one direction. The rendering came back as a roughly 12x12 square with the stairs going the wrong way. It's not really showing my deck — it's generating something else.

I'd love for the AI to either read the dimensions from the image I upload, or let me confirm the size and orientation upfront so the rendering matches the project I'm going to build. Right now I can't show these to clients because they don't match the real deck.