Hi everyone,
I’m working on an AI fashion try-on mobile app and have been testing the Nano Banana model for outfit generation.
So far, the results are not fully stable — in about 5 out of 10 tests, the model successfully dresses the person with both the top and bottom garments correctly.
In the other cases, it either keeps parts of the old clothes or layers the new ones incorrectly.
My goal is to achieve more consistent results (8–9 out of 10) with a single prompt for the mobile app.
Has anyone here managed to get more stable outputs or found prompt structures that improve accuracy?
Any API tips or best practices would be really appreciated!
