You can do it, but it takes two steps. Code is generally better to create such strict geometry (even from ambiguous prompts), while the image diffusion model is great for tuning style and lighting.

https://chatgpt.com/share/69e88b5c-8628-83eb-8851-f587ef2c95...