A loan screen builded with Gemini Pro 3

Following my last post on UI and micro-interaction tests with Gemini Pro, I pushed things a bit further.

I took an old product screen and asked the model to suggest improvements in usability, UI and motion.

Usability was the strongest part. The suggestions came clean, objective and improved the flow without losing the original intent.

In UI, things required more manual refinement. Gemini respects exactly what you ask for and doesn’t break the entire composition, but the pixel-perfect layer still depends on human hands.

For micro-interactions, it keeps surprising me. It understands the Bezier curves I define and adapts the animation style well enough to speed up iteration.

The biggest challenge was images and textures. Nothing really worked. I ended up building an entire coin illustration in code: circle, gradients generated in Figma, extrusion, embossed symbol, lighting and shadows. All step by step, because Gemini still struggles with ultra-realistic details.

Here’s the prototype if you want to explore it: https://lnkd.in/dqqWyjRc