Anyone tested Gemini 3's image-to-visualization path for math problems yet?

So I've been poking around with Gemini 3's new deep thinking mode and the multimodal stuff is actually pretty wild. Took a screenshot of a calculus problem from an old textbook, fed it in, and instead of just spitting out the answer it generated this whole visual breakdown of the problem space. No prompting for visualization, it just decided that's how it wanted to explain the solution path. The challenge is figuring out if this is consistently useful or just impressive when it works.

What's interesting is how different this feels from tools like Perplexity which are great at research synthesis but don't really do this kind of zero-shot visual reasoning. I tried the same image through Perplexity and got a solid text explanation but nothing visual. With Gemini 3 you're getting graphs, coordinate planes, function transformations without asking. It's reading the math notation from the image and building out representations that show the concepts rather than just solving algebraically. From what I've seen in my testing the accuracy on OCR of mathematical notation is finally good enough that this becomes practical.

The part that has me curious is whether this translates to user research scenarios at all. Been thinking about feeding it user journey maps or wireframe sketches to see if it can generate different visual interpretations of the same data. Ideogram does well with creating visuals from text prompts but this reverse direction where you go image to different image representations feels new. Could be useful for rapid prototyping of data viz concepts when you have rough sketches.

Has anyone tried pushing this with more complex diagram types? Wondering if it handles statistical plots or network diagrams with the same capability it shows for geometric math problems.