Another Nano banana demo for pose control !

Hey everyone!
I’ve been experimenting with Google’s Nano Banana model (Gemini 2.5 flash-image) and wanted to share a small demo of what it can achieve for pose-driven image editing. I included a before and after so you can get a clear idea of how it behaves across different pose references.

What really stood out during my tests is how well Nano Banana adapts the subject to a target pose whenever the background stays uncomplicated. Instead of simply copying the reference, the model seems to reason about the body structure and adjusts limbs and proportions in a way that feels intentional and consistent. You can give it almost anything as a guide, even a plain T-pose dummy, and it still manages to reshape the character with solid coherence.

In scenes with more visual clutter, the model still shifts the subject toward the pose you want, but the match becomes a bit looser and some details are harder for it to pin down. It keeps the overall direction, just with less precision than in clean setups.

This demo is also a small preview of a feature I’m integrating into my upcoming app, CosplayAI. If you’d like to get early access when the first version is out, you can sign up for the waitlist: https://orio2n.github.io/cosplayai.app/