But Gemini 3 introduces what Google calls “generative interfaces,” which allow the model to make its own choices about what kind of output fits the prompt best, assembling visual layouts and dynamic views on its own instead of returning a block of text.