Are there general LLMs that can translate text 'on image', as opposed to spitting out the translation as text? I feel like it would be easier to reason about in images that are very text-heavy like some of these diagrams, but from what I remember, last time I tried, chatgpt and claude would only give me a text translation