Nano Banana Pro Prompts: A Template and Six Examples
Google's Nano Banana Pro follows plain sentences about the subject, the layout, the light and any words that must appear in the image. This guide gives you one template, the edits that work with reference images and six examples to paste into Revana, where Nano Banana Pro (nano-banana-pro) costs $0.04 an image at 1K and 2K and $0.08 at 4K (prices as of November 2026). Every example runs on the same API key as the rest of our image and video models: change the model name and the same prompt runs elsewhere too.
The template that works for most images
Write it like a brief, not a list of adjectives. Name the subject first, then say where it sits and how it is lit, then add any words that must appear and a style if it is not a photo.
[Subject: what the image is about, with one or two defining details].
[Layout and camera: where it sits in the frame, the angle, a lens for photos].
[Setting and light: the place and the light source together].
[Text, if any: the exact words in "quotes", where they go, the lettering].
[Style, only if it is not a photo: one style].| Teil | What to write | Beispiel |
|---|---|---|
| Motiv | What the image is about, with one or two concrete details. Put it in the first sentence. | A matte black coffee grinder with a walnut lid |
| Layout and camera | Where things go in the frame and the view. A lens hint helps photoreal shots. | Centered, three-quarter view, 85mm lens look |
| Setting and light | The place and the light source in one phrase. Light does more for realism than quality words. | On a plain white backdrop, soft studio light from the left |
| Text and style | Quoted words with a position and a lettering style; one style line when it is not a photo. | The label reads "Morning Grind" in a bold serif |
Start with the subject
Name what the image is about and give it one or two defining details, such as a matte black coffee grinder with a walnut lid. The model holds onto concrete details better than vague words like beautiful, and everything after the first sentence describes where the subject sits and how it looks.
Set the layout and camera
Say where things go in the frame: centered, on the left third, in the foreground. Add a camera view such as three-quarter, top down or eye level when the angle matters. A lens hint like 85mm portrait or wide angle helps with photoreal shots; leave it out for illustrations and flat design.
Describe the setting and light
Name the place and the light source together, such as a kitchen counter in soft morning window light. For product shots, a plain white or grey backdrop keeps the focus on the item; for lifestyle shots, describe one or two props rather than a busy scene.
Name the style only when it is not a photo
Nano Banana Pro defaults to a realistic look, so a style line is only needed for something else: flat vector illustration, watercolor, screen print or 3D clay render. Keep it to one style per prompt; mixing three usually blends them into something none of them look like.
Getting words in the image spelled right
Quote the exact words
Put every word that must appear in quotation marks, such as a headline that reads "Open Late Fridays". Unquoted words get treated as description, not as text to draw. Short lines of two to six words come out cleaner than full sentences.
Say where the text goes
Give each line a place: across the top, centered on the label, along the bottom edge. For several labels, list them in reading order, which keeps step numbers and arrows in sequence on diagrams and menus.
Describe the lettering
Name the type style in plain words, such as bold condensed sans-serif, handwritten chalk or an old-newspaper serif, and add a color when the text must match a brand. Keep text large enough to read at the size you will use the image.
| Vague | Quoted and positioned |
|---|---|
| A cafe sign that says open late on fridays | A chalkboard cafe sign. The top line reads "Open Late" in large handwritten chalk letters, and below it "Fridays until 11" in a smaller print style. |
Editing and combining your own images
Number the images you send
Nano Banana Pro on Revana accepts up to 14 input images in one request, sent as image_urls. Refer to them by order in the prompt, such as the jacket in image 1 and the model in image 2, so the model knows which picture supplies the product, the person or the background.
Say what must stay the same
End every edit prompt with what to keep: the face, the pose, the camera angle, the background. Edits go wrong most often when the prompt only says what to change. Use photos you own or have permission to use for people and products; clear, well-lit sources give cleaner edits.
Change one thing at a time
Ask for one change per request, such as a new background, then a new color, then a new prop. Each pass costs $0.04 at 1K, and a single change is easier to judge. When an edit lands, use that output as the next input: building in steps keeps faces and products steady across a series.
Replace the background in image 1 with a sunlit concrete wall. Keep the product, its shadow and the camera angle unchanged.
Change the jacket in image 1 to deep forest green. Keep the fabric texture, the fit and everything else the same.
Add a small potted cactus on the desk to the right of the laptop in image 1. Match the window light. Keep everything else unchanged.Pick the size and shape before you write
Nano Banana Pro takes 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 5:4, 4:5 and 21:9 on Revana. Pick the ratio for where the image will live, such as 9:16 for stories or 4:5 for feeds, and set it in the request as aspect_ratio rather than writing it into the prompt, so the layout words fit the frame you actually get.
| Auflösung | Pro Bild | 100 Bilder |
|---|---|---|
| 1K | $0.04 | $4.00 |
| 2K | $0.04 | $4.00 |
| 4K | $0.08 | $8.00 |
Draft at 1K while you tune the wording, then render the keepers at 4K; the prompt does not need to change between sizes. For rough drafts, Nano Banana 2 costs $0.03 an image at 1K and 2K on Revana (prices as of November 2026) and takes the same reference images: switching is one word in the model field. The Nano Banana Pro price calculator totals a batch.
Six prompts to copy
Each one follows the template. The ones with reference images expect the pictures in image_urls, in the order the prompt numbers them.
Product shot on a clean background
A matte black coffee grinder with a walnut lid on a plain white backdrop, three-quarter view, soft studio light from the left with a gentle shadow. Photoreal, 85mm lens look, square frame.The subject comes first, then the angle, then the light. Swap the product and keep the rest for a matching catalog set.
Poster with a headline
A retro travel poster of a lighthouse on a rocky coast at sunset, flat screen-print style in teal and orange. The headline reads "Visit Cape Argus" in bold condensed letters across the top.Quoting the headline and giving it a position keeps the words clean. One style line is enough to set the whole look.
Labeled infographic
A clean infographic of the four steps of pour-over coffee, laid out left to right with simple line icons on a cream background. Label the steps "Grind", "Rinse", "Bloom" and "Pour" in a friendly sans-serif font.Listing the labels in reading order keeps the steps in sequence. Short labels read better than full sentences under each icon.
Outfit swap from two photos
Put the jacket from image 1 on the woman in image 2, keeping her face, pose and the street background unchanged. Match the light direction and add a soft natural shadow under the collar.Numbered images and a clear keep list do most of the work here. Use photos you own or have permission to use.
Character sheet
A character sheet of the girl in image 1 in a flat cartoon style, shown from the front, side and back on a plain grey background. Keep her yellow raincoat, short black hair and red boots the same in every view.Restating the defining details holds the character steady across views. Use the sheet as a reference image for later scenes.
Room restyle
Restyle the living room in image 1 as Scandinavian minimal, with light oak floors, white walls and a linen sofa. Keep the windows, the doorway and the camera angle exactly where they are.Naming the fixed parts stops the model from rebuilding the room. Try the same photo in two or three styles before you pick one.
Habits that make results worse
| Habit | Weak prompt | Better prompt |
|---|---|---|
| Stacks of quality words | portrait of a chef, 8k, masterpiece, ultra detailed, best quality, trending | A chef in a white jacket plating a dessert, eye level, warm light from the pass, shallow depth of field. |
| Telling it what not to show | A beach at dawn, no people, no boats | An empty beach at dawn, smooth wet sand and a calm sea to the horizon. |
| Asking for too much in one image | A market with ten stalls, a juggler, a dog, balloons, three signs, a band and fireworks, in watercolor and neon and pixel art | A morning flower stall in a busy market, the florist handing a bouquet to a customer, soft watercolor. |
Stacks of quality words
Words like 8k and masterpiece add little to a modern model. Describe the light, the lens and the material instead, and cut any word that does not change the picture.
Telling it what not to show
A line like no people can still put people in the frame. Describe what is there instead; in edits, say what to keep rather than what to avoid.
Asking for too much in one image
Ten objects, three text blocks and two styles crowd each other out. Pick the two or three things that matter and build the rest with edits, one step at a time.
Run a prompt through the API or your agent
Send the request
POST the model name nano-banana-pro and an input with your prompt, aspect ratio and resolution to https://api.revana.si/v1/generations. Add image_urls when you are editing or combining pictures. Poll the job or wait for a signed webhook, then download the file; the response shows the exact charge in cost_usd.
# One Idempotency-Key per job, made once: resend with the same $KEY after a timeout, a 429 or a 5xx
KEY=$(uuidgen)
curl https://api.revana.si/v1/generations \
-H "Authorization: Bearer $REVANA_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $KEY" \
-d '{
"model": "nano-banana-pro",
"input": {
"prompt": "A retro travel poster of a lighthouse on a rocky coast at sunset, flat screen-print style in teal and orange. The headline reads \"Visit Cape Argus\" in bold condensed letters across the top.",
"aspect_ratio": "2:3",
"resolution": "2K"
},
"max_cost_usd": 0.04
}'{
"model": "nano-banana-pro",
"input": {
"prompt": "Put the jacket from image 1 on the woman in image 2, keeping her face, pose and the street background unchanged. Match the light direction and add a soft natural shadow under the collar.",
"image_urls": [
"https://example.com/jacket.jpg",
"https://example.com/woman.jpg"
],
"aspect_ratio": "4:5",
"resolution": "2K"
},
"max_cost_usd": 0.04
}{
"id": "gen_...",
"status": "succeeded",
"model": "nano-banana-pro",
"output": {
"urls": [
"https://cdn.revana.si/g/gen_.../0.png"
]
},
"cost_usd": 0.04
}Quote it first
Send the same body to /v1/generations/quote and Revana returns the price without spending anything; quotes are free, so check a batch before you run it. Set max_cost_usd on the real request as a ceiling: a job that would cost more is refused instead of charged.
Ask from Claude or ChatGPT
Connect the Revana MCP server to Claude, ChatGPT, Cursor or VS Code, paste a prompt from this page and ask the agent to quote and make it. The agent runs on your prepaid Revana credit and returns the image in the chat, with no code for one-off images.
Failed images cost nothing
Revana reaches Nano Banana Pro through 3 routes and moves to the next one automatically if a route fails. A failed attempt costs $0.00: you are charged once, for the image you get back, so batch runs stay predictable even when a provider has a bad hour.
FAQ
How long should Nano Banana Pro prompts be?
Most good prompts run 30 to 80 words: a sentence for the subject, one for layout and light, and one for any text or style. Longer prompts help only when every extra word changes the picture.
Can Nano Banana Pro put readable text in an image?
Yes, it is one of the model's strengths when the prompt quotes the exact words and says where they go. Keep each line short, name the lettering style and render at 2K or 4K for small text.
Wie viele Referenzbilder kann ich senden?
On Revana you can send up to 14 input images with one Nano Banana Pro request. Refer to them by number in the prompt so the model knows which picture supplies the product, the person or the scene.
How much does each image cost?
On Revana, Nano Banana Pro costs $0.04 an image at 1K and 2K and $0.08 at 4K (prices as of November 2026). You pay from prepaid credit that never expires, with no subscription, and failed images cost $0.
Do these prompts work on other image models?
The template carries over to GPT Image, Seedream and FLUX because it describes the picture in plain sentences. Results differ by model, so run the same prompt on two or three through one Revana key and compare.
Darf ich die Bilder kommerziell nutzen?
Commercial use depends on Google's terms for Nano Banana Pro, so check them before you publish. Revana gives you access to the model and returns the files you generate, and edits should use images you own or have permission to use.
Run these prompts on your own key
Create a free Revana key, paste a prompt and get the image back for $0.04. The same key runs every other image and video model.