I built a 3D model and exported this image of it, then uploaded it to Gemini and asked it to do the work of rendering it with embellishments. The initial output had horizontal turbines on the curved edges of the vertical turbines. They have the ability to mark up the image and add instructions for modifications. It's getting better but still making some odd interpretations. Within a couple of years, this is going to be the primary tool of fantasy digital content.
Despite uploading multiple images of the 3D models I've built, #GeminiAI insists on making its own aesthetic choices, especially when I mark it up with corrections. It doesn't handle specific details nearly as well as generalized requests. The same model of a spaceship is now rendered slightly differently despite zero instructions on doing so. LOLOL
I'm still confident that, in a couple of years, one can upload images of a model and have Gemini remember it and reuse the same model.
Spin D'Accord