Gemini 2.5 is surprisingly good at image translations. I've used it a few times for untranslated booru images or background stuff in manga panels that wasn't translated, like newspapers, messages or signs.
But doing this for a VN or even a normal novel sounds pretty bad. There is only so much subtext, context and general awareness you can cram into that limited input context. And even if input wasn't a limit, I don't trust the AI's output quality. chatting with it is fine, and if you beat it hard enough it can do good smut, but that's it.