The Gemini API is Google's way of letting you call its models directly: text models in several sizes, vision on images and video, and image generation. You get a key from AI Studio, install a small SDK and start sending requests, no cloud console required for the simple cases.

How I use it

Two scripts, both small. The first one is a pair of eyes. Give the Flash model an image and a question, ask for JSON back, and you get a reliable yes or no, a list of what is in the picture, or a flag on something that should not be there, for a fraction of a cent each. I run that over folders of images before a human ever looks at them, and it catches the obvious things a tired person misses. The second script generates images. The image model renders a scene from a prompt, and unlike most, it can put a short line of text into the picture and spell it correctly. I use it for drafts, mockups and the kind of background a stock library would charge for.

What surprised me is the price. A thousand vision checks cost less than a coffee, and the free tier covered all the testing. What annoys me is the churn. Model names change, the old ones are retired with little notice, and a script that ran in spring needs a new name in autumn. The free tier's rate limits hit the moment you loop over a folder. Every generated image carries a small sparkle mark in a corner that has to be cleaned before use. And the safety filter sometimes refuses a perfectly ordinary product photo and tells you nothing useful about why.

Who it is for

Developers who want cheap vision or image generation in a pipeline and are fine with Google's pace of change. Skip it if you need a model that stays the same for a year, or if your use case is long-form reasoning; there are better choices for that, and they are listed on this page.

Related tools