I’m having a hard time imagining how failure to see an image would result in such a misleadingly specific wrong output instead of e.g. “nothing” or “it’s nonsense with no significant visual interpretation”. That sounds awful to work with.

LLMs have a very hard time saying "I am useless in this situation", because they are explicitly trained to be a helpful assistant.

So instead of saying "I can't help you with this picture", the thing hallucinates something.

That is the expected behavior by now. Not hard to imagine at all.

No controls in the training data?

Fun fact,you can prompt the llm's with no input and random nonsense will come out of them

And if you set the temperature to zero, you'll get the same output every time!