Openrouter definitely supports vision models. Why would you have to give up vision?
For example if I want to paste a screenshot of what I mean, I can't.