Nano Banana Pro Prompts for Photo Editing: 15 Examples That Work
Nano Banana Pro prompts for photo editing are plain-text instructions that change a photograph: replace the background, the outfit, the lighting or the facial expression, or build a new scene from several shots. The Google Gemini-based model reads these instructions much like a retoucher would, so the trick is to name what to change and what to leave untouched. In IIshki Studio, Nano Banana Pro takes up to 8 references and outputs up to 4K; Nano Banana 2 covers quick drafts, and GPT Image 2.5 Flare handles text on images. Below: the structure of a working prompt, 15 ready examples for common tasks, and a table of mistakes.
What Nano Banana Pro can do in IIshki Studio
Before the prompts, the limits the model works within here (taken from the model card, not from memory):
- References: up to 8 images per generation, JPEG / PNG / WebP, up to 20 MB each. A reference is optional — you can generate from scratch.
- Resolution: 1K, 2K (default) or 4K. 2K is enough for social media; 4K is for print and for images you plan to crop later.
- Aspect ratio: ten options — square 1:1, landscape 16:9, 21:9, 4:3, 3:2, 5:4, and portrait 9:16, 3:4, 2:3, 4:5. The ratio is a setting, not a line in the prompt.
- Billing: tokens per generation; the price is shown on the Nano Banana Pro card. The result is saved to your library and sent as a file to Telegram if your account is linked.
The key property for editing: the model keeps what you asked it to keep and redraws what you didn't mention. That's why half of a good prompt is a “do not touch” list.
Which model to pick for photo editing
| Task | Model | Why |
|---|---|---|
| Replace background, outfit, lighting or expression on a photo | Nano Banana Pro | Most precise reference-based editing, up to 8 images, up to 4K |
| Quickly try 5–10 edit variants | Nano Banana 2 | Same prompt logic, faster and cheaper per image |
| Insert or replace text on a photo | GPT Image 2.5 Flare | Better at rendering text and typefaces |
| Build a scene from 9–10 references, art or cartoon style | Seedream 5.0 Pro | Up to 10 references, holds stylization well |
| Upscale without redrawing | Topaz Image Upscale | Up to 4×, cleans noise, leaves content untouched |
If the job is “fix one photo”, start with Nano Banana Pro. See the head-to-head with OpenAI's model at GPT Image 2 vs Nano Banana Pro and with BytePlus at Seedream 5 vs Nano Banana Pro.
How to write a Nano Banana Pro prompt: 5 parts
The model doesn't need flowery adjectives — it needs a plan. A working editing prompt has five parts, and the order matters:
- What to change. One action in one sentence: “replace the background with…”, “dress the person in…”, “remove…”.
- What to keep. An explicit list: face, pose, hair, gaze direction, light on the skin. Without it the model will “improve” things you never asked about.
- Light and camera. Where the light comes from, warm or cool, whether the new object casts a shadow. Lighting is what gives away a bad composite.
- Style and materials. Photorealism or illustration, fabric texture, metal sheen, film grain.
- Reference roles (when there's more than one). “First photo — the person, second — only the colour and cut of the outfit, third — background and lighting.”
Don't write the aspect ratio or resolution into the text — they're chosen in the settings under the prompt field.
A prompt assembled from this scheme:
Replace the background with a café terrace in Lisbon at sunset. Keep the face, pose, hairstyle and glasses unchanged. Warm light, low sun from the left, a soft shadow on the shoulder. Photorealistic, barely visible film grain.
How to edit a photo in Nano Banana Pro, step by step
1. Prepare the source
Upload the original, not a re-compressed copy from a messenger: the model treats compression artifacts as texture and amplifies them. If the photo is small or noisy, run it through Topaz Image Upscale first — a clean frame is easier to edit.
2. Upload references and assign roles
On the Nano Banana Pro card, add one to eight images. The upload order is the numbering you refer to in the prompt (“in the first photo”, “the outfit from the second”).
3. Write the prompt: change → keep → light → style
One edit per generation. If you need a new background, a new outfit and a new expression, do it in three passes: upload the result of each step as the reference for the next. Every stage stays under control and the face doesn't drift.
4. Choose the ratio and resolution
Stories — 9:16; product card — 3:4 or 1:1; cover — 16:9 or 21:9. Start at 2K: once you like a variant, rerun the same prompt at 4K.
5. Check the result and refine
Look at the three places where image models slip most often: hands, hair edges against the new background, and shadow direction. If something is off, don't rewrite the prompt from scratch — add one clarifying phrase and run again.
15 Nano Banana Pro prompts for photo editing
Every example assumes an uploaded photo. Aspect ratio and resolution are set in the settings — no need to repeat them in the prompt.
Background and location
Background swap
Move the person from the photo into a Scandinavian interior: white walls, wooden floor, a large window on the right. Keep the pose, face, outfit and gaze direction. Cool daylight from the window. Add a soft shadow on the floor.
Time of day
Turn this daytime street shot into evening: sunset, long shadows, warm sun behind the subject. People, buildings and composition unchanged — only the lighting and sky.
Outpainting
Extend the image on both sides to a landscape format, continuing the street and sky in the same style and lighting. Do not alter the central part.
The person
Outfit
Dress the person in a black cashmere sweater and light trousers. Loose cut, visible fabric texture. Face, hair, pose and background unchanged; light on the clothes matches the light on the face.
Expression
Change only the facial expression: a calm, slight smile, eyes a little narrowed. Pose, head angle, hair, outfit and background stay exactly the same.
Hairstyle
Make four versions of the same portrait with different hairstyles — short bob, long waves, high bun, pixie cut — in one 2×2 grid. Face and lighting identical in all four.
Studio portrait
Turn this selfie into a black-and-white studio portrait: dark backdrop, one hard light source from the left, deep shadows, film grain. Facial features unchanged.
Objects
Remove clutter
Remove the people in the background and the wires above the street. Rebuild the background so no trace of the edit remains; buildings, road and lighting as in the original.
Add an object
Place an open cloth-bound book on the table next to the cup. The book's shadow falls the same way as the cup's, warm light from the window. Change nothing else.
Replace an object
Replace the bicycle by the wall with a red scooter of the same size. Wall, pavement, lighting and shadows unchanged.
Style
Vintage frame
Turn the photo into a 1970s colour-film shot: faded warm tones, light grain, soft vignette. Faces and composition preserved, clothing styled to the era.
Four seasons
Show this landscape as four vertical strips from left to right — spring, summer, autumn, winter. Same angle and composition; only vegetation, light and sky change.
Product and text
Product card
Cut out the sneakers from the photo and place them on a clean light-grey background with a soft shadow underneath. Shape, colour and lacing exactly as in the original, no new details. Even studio lighting.
Text on an object — better done in GPT Image 2.5 Flare:
Replace the text on the sign with “Bakery No. 7”, keeping the typeface, letter colour and the rust on the metal. Leave the rest of the frame unchanged.
Multiple references
Scene assembly
First photo — the person, keep their face and build. Second — only the cut and colour of the coat. Third — background and lighting. Compose a photorealistic frame: the person wearing this coat standing on this background, with the light from the third photo falling on the face and fabric.
Working with several references
Nano Banana Pro takes up to 8 images, but the more you add, the more important it is to explain what each one is for. The working scheme is one role per reference:
- Person — face, build; “keep the features”.
- Outfit — “only the cut and colour, don't take the person from this photo”.
- Pose — “repeat the body and hand position”.
- Background and light — “location and light direction from here”.
When two references disagree (light from the left in one, from the right in another), say which one wins: “lighting per the third photo”. If you need more than eight references or the job drifts into illustration, switch to Seedream 5.0 Pro; and a chain like “edit → edit → upscale” is easiest to build on the Canvas, where one model's output becomes the next model's input.
Common mistakes and how to fix them
| Mistake | Cause | Fix |
|---|---|---|
| The face became “a different person” | No ban on face changes in the prompt | Add “face, features and expression unchanged”; edit the face in a separate pass |
| New background looks pasted on, subject looks cut out | Light and shadows not described | State light direction and temperature, ask for a shadow on the floor/wall |
| Outfit from the reference arrived with someone else's face | The reference has no role | “From the second photo — only the cut and colour of the outfit” |
| Removed object left a blurry patch | Model didn't know what to fill in | Describe what belongs there: “continue the wall and tiles” |
| Hands and fingers distorted | Too many edits in one pass | Split into steps; in the last one add “hands as in the original” |
| Image is soft, details lost | Source is small or over-compressed | Upload the original; upscale first if needed |
| Text on the image has typos | Wrong model for text | Do lettering in GPT Image 2.5 Flare |
| Aspect ratio isn't what the prompt asked for | Ratio in the text is ignored | Choose the aspect ratio in the settings |
What's next
The prompts above cover 90% of everyday jobs: background, outfit, lighting, cleanup, product cards. From here it depends on the task: the round-up of models for AI photo editing, the guide to an AI photoshoot, and if the edited portrait needs to move — the article on how to animate a photo with AI. You can open Nano Banana Pro in the Studio without a subscription: the price per image is on the card, and you only pay for the generation itself.
Models in this article
Frequently asked questions
How do I edit a photo with Nano Banana Pro?
Upload the photo as a reference in IIshki Studio, write a prompt that says what to change and what to keep (“replace the background with …, keep the face, pose and lighting”), pick the aspect ratio and 1K–4K resolution, and run it. The result lands in your library and in Telegram.
Why does Nano Banana Pro change the face when editing?
Usually because the prompt never forbade it. The model redraws whatever you did not tell it to preserve. Add “face, features and expression unchanged”, upload the uncompressed original, and don't ask for a new angle and new lighting in the same pass.
How many reference images does Nano Banana Pro accept?
Up to 8 images per generation — JPEG, PNG or WebP, up to 20 MB each. Give each one a role in the prompt: first photo is the person, second is the outfit, third is the background. If you need more than ten references, use Seedream 5.0 Pro.
Nano Banana Pro or Nano Banana 2 for editing?
Pro for precise photo edits: background, outfit, lighting, multi-reference scenes. Nano Banana 2 follows the same prompt logic but is faster and cheaper per image — good for churning through many variants with fewer details.
Is Nano Banana Pro free to use?
In IIshki Studio, sign-up and uploads are free; you only pay for the generation itself, in tokens. The price per image is shown on the model card, and new club members get starter tokens for their first generations.
Every model in this roundup is available in the IIshki studio at reduced club prices. Club members get tokens for generations every month.
Join the clubMore articles
- How to Dub a Video With AI in 2026: Translate, Voice, Lip-SyncHow to dub a video with AI: translate the script, voice it in ElevenLabs or MiniMax, and lip-sync the speaker in Kling Avatars v2. Step-by-step, with prompts.
- Kling Avatar: How to Make a Talking Photo with AI in 2026How to make a talking photo with AI: portrait plus voice in Kling Avatars v2, text-to-speech in ElevenLabs and MiniMax, in-scene dialogue with Hailuo H3.
- 20 Kling 3.0 Prompts for AI Video: Ready-to-Use ExamplesReady-to-use Kling 3.0 prompts for portraits, products, landscapes, multi-shot and sound, plus the prompt formula and common mistakes. Run them in IIshki.