What’s new
- Its official developer/API name is Gemini 3 Pro Image: Nano Banana Pro is what it's called in the Gemini app and other consumer products. Built on Gemini 3 Pro's reasoning, Google positions it as its best model for complex, multi-turn generation and editing.
- Text rendering is a strength: Google calls it the model that "creates images with correctly rendered and legible text directly in the image," in multiple languages, and it can translate designs for different locales.
- Studio-level control: camera angle, shot type, color grading and lighting direction can all be set individually; across multi-turn edits it can also shift depth of field to refocus on a different subject.
- Keeps up to 5 characters and 14 objects consistent, at resolution up to 4K: every output carries an invisible SynthID watermark that can identify it as AI-generated or -edited.
What it’s good for
- Polished shots that need precise camera control — ad visuals and poster design where angle, lighting and color grading matter.
- Anything that needs accurate on-image text — posters, charts, multilingual material — with clean text rendering and localization.
- Complex, multi-round edits, like shifting depth of field to refocus the shot while keeping people and props consistent across versions.
Prompting tips
- Cover five elements: style, subject, setting, action and composition — the more detail you give, the closer the result gets to what you pictured.
- Write studio-control language straight into the prompt: camera angle (high/low/eye-level), shot size (wide/medium/close-up), lighting direction and color grading can all be specified directly.
- Use one of the five official editing moves: swap the character, adjust the composition, change the action, swap the setting, or rethink the style — pick one at a time and describe exactly what should change.
- Put on-image text in quotes: quote the words you want rendered, and note the typography and language.
- Name your characters and objects when using reference images: upload clear references and give people or objects consistent names so they stay recognizable across generations.
- Iterate one step at a time across multiple edits: feed the previous version back in as input, request a single change, and state what needs to stay the same.
Product advertising photography, a bottle of amber perfume on a dark marble counter, 45-degree
backlight, warm color grading, shallow depth of field, blurred background. Composition: medium
close-up, slightly low angle. The label reads "AMBER NOIR" in a clean serif font. 16:9, 4K.
Known limits
- Small faces, exact spelling and fine detail can still go wrong: Google acknowledges these issues exist.
- Its world knowledge is broad but not infallible: Google says its real-world knowledge is "extensive but not infallible," and localized translation can also get grammar, spelling, cultural nuance or idiom wrong.
- Advanced edits and character consistency aren't guaranteed every time: for masked editing or major lighting changes, Google states it "may not always get it right."
Using it in Nomi
On the generation canvas, add a image card, pick “Gemini 3 Pro Image” in the model picker and choose a mode (Text to image, Image to image). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
