Image generators are good at making a subject, but they do not always return the clean transparent background a design workflow needs. NanoBanana PNG tries a simple workaround: ask Gemini for the subject on a solid green background, then remove green pixels and return a PNG with alpha transparency.
The service wraps that flow in a small FastAPI endpoint. It is a useful example of combining an image model with ordinary image processing: the model creates the picture, and Python's imaging tools handle the predictable background cleanup.
Code: nanobanana-png.