Generate images from text prompts using the NanoBanana v2 model; supports an optional seed for reproducible outputs.