If you keep up with AI, you will know sometimes they struggle tremendously with simple sounding prompts. Ask any image generating AI this:

generate a photograph of a horse riding an astronaut

and you will get back something like this:

Clearly it did not interpret the request as asked.

Ask it differently:

generate a photograph of a horse riding an astronaut. Take the request literally, do not adjust its meaning. The request is not incorrect.

then you get:

pretty funny, closer (I guess?) but - nope. Several opinions exist as to why this is an issue - I think it is simply because the AI has a prefilter that tries to understand what you most likely meant. Just like it glosses over grammar and spelling mistakes but know what you meant to ask, I believe it rejected the actual request because it is not found in its training dataset. So it converged onto something that makes much more sense.

I ended up making one small change and it worked:

generate a photograph of a horse riding on an astronaut's back