How the AI image generator works
You write a description, we send it to the nano-banana-2-lite diffusion model, and the model returns a finished image. Everything happens server-side, so your browser never handles keys or captchas. Generation runs in two steps — the job is submitted first, then the page polls until the picture is ready — which is why the progress bar keeps moving instead of the page hanging.
Write prompts that actually work
- Subject first. Say what the image is of before anything else: “an old fisherman mending a net”.
- Add the setting. Time of day, weather, location: “on a wooden pier at sunrise, light fog”.
- Name the style. Photorealistic, watercolour, 3D render, oil painting, anime, isometric.
- Give camera hints. “85mm lens, shallow depth of field, cinematic lighting” pushes it toward photography.
- Keep it under ~60 words. Long prompts dilute the important parts.
Editing an existing image
Paste a public https image URL in the reference field and describe the change — “make it look like a watercolour painting” or “turn this into a cyberpunk street”. The model uses your image as the starting point instead of a blank canvas.
Good to know
- Only
autoand1:1ratios are supported by this model. - Images are hosted by the model provider — download the ones you want to keep.
- The tool is rate-limited and site-only; for programmatic access use the developer API.
- Do not generate illegal content, or images of real people intended to deceive.