Google released Nano Banana Pro on November 20, a new version of its popular image generation and editing model, and one of its headline claims addresses a problem that has dogged AI image tools since they appeared: writing text that people can actually read. Google says the model, formally called Gemini 3 Pro Image, can generate images with accurate text in multiple languages, making it suitable for posters, mockups, advertisements and other content aimed at international audiences.
The original Nano Banana, released earlier in 2025 as Gemini 2.5 Flash Image, became widely used for its ability to edit photographs while keeping faces and objects consistent. Text, however, remained unreliable, especially outside English.
Why text in images was hard
Image models have historically treated letters as shapes rather than language, producing strings that look like writing from a distance but dissolve into nonsense up close. The problem is worse for scripts with complex rules. Languages written right to left, such as Arabic, Urdu and Persian, use letterforms that change shape depending on their position in a word and connect to their neighbors. Earlier models often produced broken, disconnected or reversed characters that were impossible to read.
Google attributes the improvement to building the model on Gemini 3 Pro, its newest general-purpose model, which gives the image generator a stronger understanding of language and world knowledge. According to Google, Nano Banana Pro can render longer passages of legible text, translate text within an image, and apply a consistent typographic style, with output at resolutions up to 4K.
What early users report
Content creators working in Urdu and other South Asian languages have reported in the days since launch that the model produces correctly spelled, properly joined text in their scripts, something they say previous tools could not do. Those reports are anecdotal and Google has not published per-language accuracy figures, but if they hold up, the change removes a step from a common workflow: designers who previously generated an image and then added non-English text in Photoshop or a similar editor can now ask for the finished graphic in one prompt.
Availability
Nano Banana Pro is available in the Gemini app, in Google Workspace products including Slides and Vids, in NotebookLM, and to developers through the Gemini API, Google AI Studio and Vertex AI. Google says images generated by the model carry its SynthID digital watermark, and the Gemini app can check whether an uploaded image was produced by a Google AI model.
