Gemini image editing capability issue

Gemini API
1

I tried using Gemini Nano Banana Pro in Make to automate image generation. The goal is to add water to a pool, but it seems to be struggling a lot. It keeps hallucinating and adding small details, even though my prompt structure is strong. It includes locking the original details, preserving them, setting failure criteria, and strict prohibitions yet it still hallucinates. I just wanted to ask whether Gemini models struggle with adding water to an image?