Nano Banana Prompt Generator: Gemini Image Briefs That Stick

A nano banana prompt generator should turn a rough idea or an uploaded photo into a Gemini image brief with five fields: subject, job line, keep list, change list, and style plus output spec. That brief holds up across edit turns, so the face, product, or layout you care about stays put while you change the rest. You leave with the brief format, two input paths (idea and photo), a three-step workflow, four paste-ready briefs, and model notes for Nano Banana 2 and Nano Banana Pro as of September 2026. Viral trend prompts live in a sibling list. Soft path: https://promptmake.net/for/nano-banana drafts the brief text. You paste it into Gemini yourself.

What a nano banana prompt generator should output

Nano Banana is the public name for Google's Gemini image models. Nano Banana 2 runs on Gemini 3.1 Flash Image and serves as the default for most image work in 2026. Nano Banana Pro runs on Gemini 3 Pro Image and handles harder jobs: dense layouts, readable lettering, and high-resolution exports. Both sit behind a Gemini reasoning model that plans the scene before it renders.

That planning step changes what a good prompt looks like. Tag soup from the Stable Diffusion era ("masterpiece, 8k, trending") gives the planner little to work with. A short brief in plain sentences gives it a subject, a job, and limits. Google's own Gemini 3 guidance asks for concise, direct instructions, and the image models follow the same rule.

A generator earns its place when it produces that brief every time, in the same field order, so you can compare takes and swap one field at a time. A one-line "make it cool" prompt gives Gemini room to redesign the face, the product label, and the background at once.

Scope note: the sibling guide at https://promptmake.net/blog/nano-banana-prompts-guide collects trend prompts such as figurine and retro mall portraits, plus tier basics. This page covers the generator workflow and brief format you can reuse for any job, trend or not.

The Gemini image brief: five fields

A Gemini image brief is a short block of plain sentences in a fixed order. The order matters because the reasoning front-end reads top to bottom and weighs early lines as the core of the job. Put the subject first so Gemini knows what the picture is about. Put the job line second so it knows whether to invent pixels or edit your upload. Then list what must stay and what must change. Close with style and the output spec. A brief under about 120 words usually beats a long one, because the planner can hold every constraint at once. Write each field on its own line when you draft, then paste the block as one message.

Subject

Name the main thing and its defining traits: "A matte black ceramic travel mug with a cork lid and a small embossed leaf logo." For people, describe role and visible traits rather than a named celebrity. The subject line should survive unchanged across edit turns.

Job line: create or edit

Start with a verb that names the job. "Create an image of" for text-only generation. "Edit this photo" or "Using the attached image" for uploads. Gemini handles both inside one model, and a clear job verb stops it from treating your reference as loose inspiration when you wanted a precise edit.

Keep list

List what must not move: "Keep the person's face, hairstyle, and glasses. Keep the mug shape and logo placement. Keep the camera angle." The keep list does most of the work on edits. Without it, Gemini may improve things you never asked it to touch.

Change list

List the changes, ideally one or two per turn: "Change the background to a sunlit kitchen counter. Change the lighting to soft morning window light from the left." Short change lists make it easy to see which instruction caused a result. Long ones hide the cause.

Style and output spec

Close with medium, light, and mood, then the output: "Commercial product photography, natural color, shallow depth of field. Square 1:1, high detail." Put any in-image text in quotes with placement: "Headline 'Fresh Start' in bold sans serif, top third, white." Nano Banana Pro renders lettering better than Nano Banana 2, so switch to Pro when text must read cleanly.

Two input paths: rough idea or photo

Generators take one of two inputs, and each path fills the brief in a different way. A rough idea starts with nothing locked, so the generator must invent the subject details, pick a style, and leave the keep list short or empty. A photo starts with a lot locked: face, product geometry, pose, and framing already exist, so the keep list gets long and the change list does the creative work. Decide which path you are on before you write anything. Mixing them, such as uploading a photo and then describing a different subject, confuses the planner and wastes a generation on a hybrid neither of you wanted.

From a rough idea

Write one sentence about the picture and its use: "Hero image for a cozy bookshop newsletter, autumn feel." Expand it into the five fields. Subject: "A small corner bookshop window with stacked hardcovers and a sleeping orange cat." Job: "Create an image of." Keep: skip on the first take. Change: skip. Style and spec: "Warm evening light, soft film look, 16:9, space on the left for a headline."

After the first take, the keep list starts to fill. If the cat and window came out right, add "Keep the cat, window frame, and book stacks" and change only the light or the season.

From a photo

Upload the photo and describe what you see before you ask for changes. That description becomes the subject and keep list. "Edit this photo of a woman in a denim jacket standing on a city sidewalk. Keep her face, hair, jacket, and pose. Change the background to a rainy night street with neon reflections. Cinematic photo, cool tones, 4:5 vertical."

When you cannot put the photo into words, run it through https://promptmake.net/image first. The Recreate mode writes a structured description you can trim into the keep list. Use photos you own or have rights to, and skip uploads of other people's faces without consent.

Step-by-step generator workflow

The workflow below takes about ten minutes the first time and two or three once you have saved briefs. It assumes the Gemini app's image tool or any surface that exposes Nano Banana 2 or Pro. Keep a notes file open beside the chat before you start. You will paste the brief there, mark which field you changed on each turn, and save the winner as a shell at the end. That record turns one good image into a repeatable process. Without it, you end up with a folder of nice pictures and no idea which words produced them.

Step 1: Draft the brief

Write the five fields in order, one line each. Read the brief out loud. If the style line comes before the subject, or the change list names five things, fix it before you paste. Pick the tier: Nano Banana 2 for most drafts, Pro for lettering or dense layouts.

Step 2: Generate once, then edit in the same thread

Paste the brief and generate. Score the take against the keep list first, then the change list. If one thing is wrong, send a short follow-up in the same chat: "Keep everything else. Make the neon reflections softer and remove the parked car." Gemini 3 image models carry visual context across turns in one thread, and a fresh chat drops that context.

Step 3: Save the brief as a reusable shell

When a take works, copy the brief into your notes and replace the specifics with brackets: [subject], [keep items], [change items], [style], [ratio]. Record the tier and date beside it. The next job starts from that shell instead of a blank box.

Paste-ready Gemini image briefs

Fill the brackets and paste each brief as one message. Keep follow-ups to one or two changes.

Brief 1, product edit: "Edit this product photo of [product and key traits]. Keep the product shape, label layout, and colors exactly as shown. Change the background to [surface and setting] with [light direction] soft light and a gentle shadow under the product. Commercial product photography, natural color, no extra props. [Ratio], high detail."

Brief 2, portrait restyle: "Using the attached portrait, keep the person's face, hairstyle, and expression. Change the outfit to [outfit] and the background to [setting]. Change the lighting to [light]. Editorial photo, realistic skin texture, single subject, no text. [Ratio]."

Brief 3, scene create from an idea: "Create an image of [subject with two or three concrete details] in [place and time of day]. [Light and mood]. [Medium: photo, watercolor, 3D render]. Leave empty space on the [side] for a headline. [Ratio]. No text, no logos."

Brief 4, poster with lettering (use Pro): "Create a poster for [event]. Main image: [subject]. Render the headline '[exact words]' in [font style] across the top third, and the line '[date and place]' small at the bottom. [Color palette]. Clean layout, high contrast, vertical 2:3."

Common mistakes with nano banana prompts from a generator

• Skipping the keep list on edits. Gemini polishes the face or label along with the background.
• Stacking five changes in one turn. You cannot tell which one broke the image.
• Starting a new chat for each edit. The model loses the visual context of earlier turns.
• Pasting keyword soup. The planner reads sentences; weight syntax and quality tags add noise.
• Using Nano Banana 2 for dense lettering. Switch to Pro when words must read cleanly.
• Treating this workflow page as a trend list. Trend prompts sit in nano-banana-prompts-guide; this page owns the brief format.
• Uploading faces or brand assets you do not have rights to. A good brief does not fix a rights problem.

Nano Banana model notes (September 2026)

Three models share the name. The original Nano Banana runs on Gemini 2.5 Flash Image and still suits cheap, quick edits. Nano Banana 2 (Gemini 3.1 Flash Image) launched in February 2026 and serves as Google's recommended default for most new image work. Nano Banana Pro (Gemini 3 Pro Image) costs more and handles strict layouts, lettering, and 2K or 4K output.

Gemini 3 image models accept several reference images per request, which helps when you need a product plus a style swatch, or a character plus a costume detail. Google's API docs list up to 14 references, and the practical limit varies by surface, so start with two or three. Search grounding can pull real references before generation on some surfaces; check any factual detail it draws, such as a chart or a landmark.

Free Gemini accounts carry daily image caps and may add a visible watermark as of mid-2026. Developers reach the same models through the Gemini API with model IDs such as gemini-3.1-flash-image. Verify current names in the model picker before you write them into a team template.

Use PromptMake /for/nano-banana to draft briefs

Paste your rough idea into https://promptmake.net/for/nano-banana and the generator returns a natural-language brief shaped for Gemini's image models: subject first, clear job, light, style, and output spec. Add your keep and change lists, then paste into Gemini. When you start from a photo, describe it with https://promptmake.net/image and fold that description into the keep list.

PromptMake writes prompt text only. It does not open your Google account or render images. As of September 2026, guests get about three free generations per day per tool and registered free accounts about five, separate from Gemini's own caps.

FAQ

What does a nano banana prompt generator do?

It turns a rough idea or a photo description into a structured prompt for Google's Gemini image models. A good one outputs the same fields each time: subject, job line, keep list, change list, and style plus output spec. You paste that brief into Gemini and iterate with short follow-ups. The generator writes text; Gemini renders the image.

Is Nano Banana the same as Gemini?

Nano Banana is the public name for Gemini's image generation and editing models. Nano Banana 2 runs on Gemini 3.1 Flash Image and Nano Banana Pro runs on Gemini 3 Pro Image. You reach them through the Gemini app, Google AI Studio, or the Gemini API. The chat models such as Gemini 3.5 Flash handle text, not pixels.

How do I keep a face or product the same across edits?

Write a keep list in each brief that names the face, hairstyle, product shape, or label layout. Change one or two things per turn and stay in the same chat thread. Gemini 3 image models carry context between turns, and a new chat loses it. If drift continues, attach the original image again with the keep list.

Should I use Nano Banana 2 or Nano Banana Pro?

Start with Nano Banana 2 for most drafts, since Google recommends it as the default and it runs faster and cheaper. Switch to Pro when the image needs readable lettering, a strict layout with many constraints, or a 4K export. Record which tier produced each saved brief. The same brief can look different on the other tier.

Can I paste Midjourney prompts into Nano Banana?

Strip the parameters first. Flags such as --ar and --sref mean nothing to Gemini, and comma tag lists read as noise to its planner. Rewrite the idea as plain sentences in the five-field order. State the aspect ratio in words, such as "vertical 4:5."

How long should a Gemini image brief be?

Aim for about 40 to 120 words. Short briefs leave room for the model to guess, while long briefs bury the keep list under style adjectives. Put each field on its own line when you draft. Cut any adjective that does not change the picture.

Is there a free nano banana prompt generator?

Yes. https://promptmake.net/for/nano-banana drafts Gemini-shaped briefs, with about three free generations per day for guests and about five for registered free accounts as of September 2026. Those quotas are separate from Gemini's image caps. Paste the output into Gemini and render there.