Nano Banana Pro Prompts and Tutorial for Sharp, Legible Images

Nano Banana Pro reads a prompt as a description of a finished picture, and it is unusually literal about placement and text. So describe the picture you would hand to a client, in plain sentences, in this order.

1. Medium: photograph, poster, menu board, product advertisement, recipe card.
2. Subject and action: a climber chalks up mid-conversation with a friend belaying below.
3. Setting and light: a climbing gym at closing time, evening light through high windows.
4. Text, in quotes, with its style: headline "SOUND, CARVED." in a clean sans-serif.
5. Layout: the headline across the top, the product centred, the list down the left.
6. Finish: candid, sharp detail, 16:9.

A photograph needs the first three and the finish; a poster needs all six. Use nouns and verbs, not adjectives: "a vendor holds up a chipped enamel teapot to a laughing couple" is a scene, where "vibrant, stunning, highly detailed" is nothing to draw.

Three models carry the name here. The table is printed from our catalog, so the sizes and the edit column are what the studio offers today.

Nano Banana Pro is the highest fidelity of the three, Nano Banana 2 is the volume model, and Nano Banana is the original. Each card below shows a still that model rendered.

• 
  GoogleNano Banana Pro1K, 2K, 4K
• 
  GoogleNano Banana 20.5K, 1K, 2K, 4K
• 
  GoogleNano Banana1K

Nano Banana Pro text rendering is the reason most people are here: a headline, a menu or a label comes back spelled as you wrote it when you follow four rules.

• Quote the exact words, in the case you want: headline text "HIYASHI SODA", not a soda headline.
• Keep each run short. A headline of two to four words and a list of five short lines render; a paragraph does not.
• Describe the type as a kind, not a font: hand-lettered chalk, refined serif, condensed sans.
• Say where it goes and how big: across the top, small legible type down the left, on the can's label.

Check the text first when the image comes back. If one letter is wrong, re-run the same prompt rather than rewording it; the studio returns up to four takes a run. The Nano Banana page has a wall of these with the prompt under each.

Nano Banana Pro for ads comes down to two patterns: a fresh render for a mock-up, where the model invents the object, and, for anything a customer will see, an edit of your own product photo, because a model asked to draw your bottle from words will draw a bottle, not yours.

• Mock-up: "A minimalist product ad for a matte black wireless speaker on a stone plinth, headline text “SOUND, CARVED.” in clean sans-serif, soft studio gradient background, premium tech-advertisement layout." This prompt rendered the product ad on our Nano Banana page.
• From a photo: attach the packshot, then "the same bottle, unchanged, on wet black slate with a single hard rim light, a thin reflection below, nothing else in frame, 4:5".
• Lifestyle: attach the photo, then "the same jar on a sunlit kitchen counter beside a cut lemon and a linen cloth, morning light, editorial food photograph, the label unchanged".

Say "unchanged" about the product every time. For a campaign, keep the first sentence identical across prompts and change only the setting. Our image to image page has before and after pairs of exactly this.

Nano Banana Pro photo editing is the same prompt box with a photo attached: upload the image in the image studio and write the change. Name what changes, name what stays, and ask for one thing per run.

For a person, write "same face, same expression, same hair" into every edit. Some jobs are a button rather than a sentence: the Editor paints a mask to retouch, extends the canvas, swaps the background, upscales, moves the camera angle and translates the text in a picture.

Nano Banana Pro character consistency is a method, not a switch: attach a reference image to every prompt and describe only the new scene, so the model edits the character into it rather than inventing a new one. The Characters library stores reference images for reuse.

• Make the character once, at 2K or 4K, in neutral light, and save that image as the reference.
• In every later prompt, attach it and start with the fixed traits: "the same woman, short grey hair, round glasses, navy coat".
• Change one thing per image: the setting, or the action, or the camera.
• Keep the medium and the finish identical across the set.
• Render four takes a run and keep the one whose face matches.

Text-only consistency works for a mascot with few features and fails for a realistic face. When the set becomes a video, our image to video page covers taking a Nano Banana Pro frame into a clip.

Nano Banana Pro renders 1K, 2K and 4K. On DreamClip 1K and 2K cost the same and only 4K costs more, so draft at 2K, where small type is easier to check, and render 4K only for the keeper that needs it. Match the shape to the placement too; a square render cropped to a story loses its headline.

Auto shape lets the model pick the frame; fine for exploring, wrong for a placement with fixed dimensions. To take a finished 2K image larger without paying for a 4K render, the Enhancer upscales it.

The first six rendered images on our Nano Banana page, on Nano Banana, and run as written on Nano Banana Pro. The seventh made the Nano Banana Pro preview above. The last five are patterns; no render is claimed for them.

1. Open the image studio on Nano Banana Pro. Its card shows where the price of an image begins.
2. Pick the shape for the placement and the size: 1K or 2K to draft, which cost the same on Pro, and 4K only for a keeper that will be printed or shown large.
3. Paste the prompt, keeping the order above. For an edit or a character, attach the photo first and describe only the change.
4. Run it. Up to four takes come back; credits settle only when they arrive, so a take that never arrives is never billed.
5. To test the same prompt on another model, open Compare, switch it to image, and run it on up to four models at once, paying only for the lanes that deliver.

For the wider field, our best AI image generators roundup ranks twelve models by job, and the image generator page shows the same prompt style across several of them.

Skin is its weak spot: faces come back smoother and more even than a camera would catch them, so for candid people Seedream 5.0 Pro is the first alternative. Its 4K costs more per image than 1K or 2K, so a set drafted at 4K pays for every miss. Each prompt is one picture, so a twelve-frame storyboard is twelve prompts. And consistency needs a reference attached every time.

For legible text, faithful layout and clean photo edits in one model, Nano Banana Pro is the one to write for, and the prompts that work are the plain ones.