How to use Nano Banana for ads: statics and start frames

What is Nano Banana, and what do creators use it for in ads?

Nano Banana is Google's name for its Gemini image models. It makes and edits images from text plus reference images. It does not make video.

Creators use it for two jobs. The first is the finished static ad: a product shot, a lifestyle scene or a text-on-image layout. The second is the start frame, a realistic still of the creator and product that a video model then animates. The tutorials split roughly evenly between the two.

Facts from Google, checked 12 Oct 2026 (Google Cloud guide, Gemini API docs):

• The models: Nano Banana 2 is Gemini 3.1 Flash Image and Nano Banana Pro is Gemini 3 Pro Image. The API docs recommend Nano Banana 2.1 for new projects.
• Reference images: a prompt takes up to 14, with per-model limits for objects and characters.
• Output size: 1K, 2K or 4K, in ratios that include 1:1, 4:5, 9:16 and 16:9.
• Watermark: every image carries a SynthID watermark.

Which tools do creators pair with Nano Banana?

About half of the tutorials stop at the still; the rest hand it to a video model. Based on 20 public tutorials posted between 1 Jan 2026 and 2 Oct 2026, reviewed by Anglr.

How do you use Nano Banana to make an ad, step by step?

The tutorials agree on starting from the real product and on generating many options before you pick. They differ on which image model to trust with text.

1. Start from the real product photo. @oggii_0 and @MayorKingAI attach the product image rather than describing it, and @oggii_0 lets the model pick props to suit the product. Google documents this case with its own formula: reference images, then a relationship instruction, then the new scenario.
2. Turn one good prompt into a template. @AmirMushich, @Kerroudjm and @ChillaiKalan__ keep the layout and lighting fixed and leave placeholders for the brand, product or ingredients.
3. Generate options, then pick and upscale. @hellorob, @Gdgtify and @gizakdag ask for a grid or a set of images in one go. Hellorob's warning: each grid cell is small, so extract the one you want and upscale it before any product text has to be read.
4. For video, make the start frame look like a phone photo. Tyler Reed, @wizzardouthere and Eugenio Zabell ask for natural light, a vertical frame and a selfie angle, with no studio look, so the video doesn't read as AI from its first frame.
5. Hand the still to a video model. Tyler Reed, Will Connell, Hamza Malik and @maxxmalist animate in Kling. @SaniBulaAI makes a first and a last frame and lets the video model move between them, which Google also suggests for Veo.
6. Plan longer ads as a set of stills. Israr Ahmad makes one still per scene of an eight-scene storyboard, @Strength04_X makes the whole storyboard on one page, and Estefania Freites builds the character sheet, locations and exact start frame. Our AI storyboards guide covers this step.

Where they differ: Will Sartorius moved handwritten formats to GPT Image but keeps Nano Banana 2 for brands whose photography looks like a retail shelf. @AIwithSynthia runs the same prompt on both models and keeps the better result, which is the cheapest way to settle it for your product.

How should you prompt Nano Banana for an ad?

Write full sentences, not keyword lists. Google's text-to-image formula is subject, action, location or context, composition, then style. Its advice: be specific, explain the purpose of the image (an ad for a skincare brand, for example), and describe what you want rather than what you don't.

For a start frame with an attached product, follow Google's reference formula. This short template is ours, built on that structure:

Use [Image 1: product photo] exactly as shown, same label and colours.
Show a [age] [person] holding it [action], in [location], [time of day] light.
Vertical 9:16 phone photo from a [selfie / eye-level] angle, natural skin,
no studio lighting. Leave the top and bottom of the frame clear of text.

Full static ad templates are on Nano Banana prompts, and the AI ad prompt builder writes a prompt in Google's order. If a JSON-style prompt is what you have been using, see JSON prompting for how it maps onto the same fields.

Common problems and fixes

Handwritten formats are the clearest split. Will Sartorius writes: "Nano Banana 2 defaults to a handwriting font, which is the wrong answer." He uses GPT Image for notes and letters.

What creators report

These are the authors' own claims, not tested by Anglr.

• Tyler Reed reports a UGC-style clip in about five minutes, from one Nano Banana Pro image and one Kling generation.
• Hamza Malik reports that his animated ads, built from Nano Banana Pro start frames, beat standard UGC on hook rate in several ad accounts.

A clean still is the start, not the ad. If your statics and AI videos already perform, Anglr pays AI creators on performance to remake winning scripts: see what we look for. For full workflows, read the AI static ads guide or browse AI ad guides.