Gemini image prompts

Nano Banana Prompts That Keep Your Face and Your Product

Nano Banana (Gemini's image model) edits your own photos when the prompt names one change and lists what must stay the same. Fill the fields, copy the prompt, paste it into Gemini.

Build a prompt See examples

Free, no sign-up

15 models, image and video

Updated October 2026

AI model What do you want?

Describe it Keep exactly the same

Setting Style

Lighting Shot / framing

Camera move Lens look Format

Exact text in the image (optional)

Sound (optional) Spoken line (optional)

Avoid (optional) Stylize --s (0-1000, optional)

Your prompt for Nano Banana

Describe what you want on the left. Your prompt appears here as you type.

Open Nano Banana

Free: 5 AI runs a day, 3 variations each. Pro: 100 runs, 10 variations, saved prompt library.

Tips

Refine in the same chat: one small change per turn ("make the light warmer, keep everything else").
Upload your photos first and refer to them as image 1, image 2 in the order you attached them.

Image to prompt

Upload a picture and the AI writes a prompt that recreates it for the model picked in the studio above, in that model's format. Your image is processed in memory and never stored.

Your image (JPEG, PNG or WebP)We shrink it to 1024 px in your browser before sending. It is not saved on our side. Note for the AI (optional)

Pro feature. Image to prompt is part of All-Access Pro: 20 images a day, plus 100 AI runs, 10 variations and a saved prompt library. Already Pro? Sign in.

Prompt from your image for the selected model

Pick a model in the studio above, upload an image and press "Write prompt from image".

Nano Banana Prompts: examples you can copy

Each example was written by the studio above. Click "Edit in the studio" to load it into the form and change the details.

Professional headshotNano Banana

Using image 1, turn the person into a professional headshot in a navy blazer over a white shirt, plain light grey studio background. Keep their face, facial features, skin tone, hairstyle and age; do not slim, smooth or beautify exactly the same. Lighting: soft key light with gentle fill. Match the lighting, perspective and grain of the original so the change looks natural. Square 1:1. Do not change anything else.

Background swapNano Banana

Using image 1, replace the background with a sunlit Scandinavian living room. Keep the subject, its edges, pose, colors and size exactly the same. Match the lighting, perspective and grain of the original so the change looks natural. Do not change anything else.

Restore an old photoNano Banana

Using image 1, restore the photo: remove scratches, dust, creases and stains, fix faded contrast, sharpen softly. Keep every face, expression and age; do not invent detail; keep it black and white exactly the same. Match the lighting, perspective and grain of the original so the change looks natural. Do not change anything else.

Product on a modelNano Banana

Using image 1, show the person from image 2 wearing the garment from image 1, full-body shot on a light grey studio background. Keep the garment's exact color, pattern and logo placement, and the person's face, body and pose exactly the same. Match the lighting, perspective and grain of the original so the change looks natural. Vertical 4:5. Do not change anything else.

Day to nightNano Banana

Using image 1, change the scene from day to night: dark blue sky, warm lights in the windows, glowing street lamps, wet reflective pavement. Keep the buildings, camera angle and composition exactly the same. Match the lighting, perspective and grain of the original so the change looks natural. Do not change anything else.

New image from scratchNano Banana

Create a studio product photo of a matte black 500 ml steel water bottle with a bamboo lid. Setting: a sage green seamless paper background. Composition: three-quarter angle, empty space on the left for a headline. Lighting: large softbox key light from the upper left, gentle rim light. Aspect ratio: Landscape 16:9.

How Nano Banana reads a prompt

It follows plain sentences and works like a conversation: upload, ask for one change, then refine in the same chat.

Refer to images by order

Upload first, then write "Using image 1" (and image 2 for a second reference). The studio writes edit prompts this way.

Always give a keep-list

"Keep the face, hairstyle and skin tone exactly the same." Without it the model may redraw everything.

Say the ratio in words

"Vertical 4:5" or "Square 1:1" inside the prompt, or pick the shape in the app. Midjourney-style --ar does nothing here.

One change per turn

Lighting now, background next. Small steps keep identity stable and you can go back one turn if a result drifts.

The Nano Banana edit formula

Using image 1, [ONE CLEAR CHANGE]. Background swap, relight, outfit, season, restore, colorize: name only one.
Keep [WHAT MUST NOT CHANGE] exactly the same. Face, pose, product shape, label text, composition, colors.
Match the lighting, perspective and grain of the original. This makes added or swapped elements look photographed, not pasted.
Do not change anything else. End every edit with it. It noticeably reduces drift.
Refine in the same chat "Make the light warmer, keep everything else." If it drifts, restart from the original photo, not the edited one.

AI image models: prompt cheat sheet

Checked October 2026. Model syntax changes often; we review this table every month.

Model	Modes	Negative prompt	Key tip
Nano Banana Google Gemini image	New image, Edit a photo	One Avoid sentence	Refine in the same chat: one small change per turn ("make the light warmer, keep everything else").
GPT Image (ChatGPT) OpenAI	New image, Edit a photo	One Avoid sentence	Ask for exact wording ("use this text exactly") when the image contains text.
Midjourney V7 Midjourney	New image, From a reference image	--no list	Keep --seed fixed while you change one word to see what that word does.
Flux Black Forest Labs	New image, Edit (Flux Kontext)	None, phrase positively	Flux ignores negative prompts: say what you want instead ("an empty street", not "no cars").
Ideogram Ideogram	New image, Edit a photo	Separate field	Turn Magic Prompt OFF when exact wording or layout matters.
Stable Diffusion Stability AI (SD 3.5, SDXL)	New image, Image to image	Separate field	Fix the seed while you test one change at a time.

Want ready prompts instead of building them?

The AI Image Prompt Pack has 64 tested prompts with formulas, a camera and style cheat sheet, troubleshooting and an assistant prompt for Claude or ChatGPT.

AI Image Prompt Pack

64 prompts as a plain-text file, yours to keep. One-time payment.

Get the pack, $9

Or everything with Pro

100 AI runs a day with 10 variations, prompts from your images, a saved library of 500 prompts with tags and search, every prompt pack, and 10 files a month.

See Pro

More prompt tools

AI image prompts Midjourney prompt generator ChatGPT image prompts AI video prompts ChatGPT photo editing prompts Gemini prompts DALL-E prompts Prompt optimizer for text AI

FAQ

What is Nano Banana?

Nano Banana is the popular name of Google's Gemini image generation and editing model. You use it in the Gemini app, in Google AI Studio or in apps built on it.

What are good Nano Banana prompts?

Full sentences with one clear change, a list of what must stay the same, the light and the format in words. The studio builds that structure from your answers.

How do I keep my face the same in Nano Banana?

Upload a clear photo, write "Keep the face, facial features, skin tone, hairstyle and age exactly the same", and change one thing per turn. If the face drifts, start again from the original photo.

Do Midjourney parameters work in Nano Banana?

No. --ar, --s and --no only work in Midjourney. In Nano Banana say the ratio in words and put exclusions in one Avoid sentence.

Is the Nano Banana prompt generator free?

Yes, unlimited. AI variations are free 5 times a day; Pro gives 100 a day with 10 variations each and a saved prompt library.

Can it write a prompt from my image?

Yes, with Pro. In Image to prompt upload a picture, pick the model in the studio, and the AI writes a prompt that recreates it in that model's format, up to 20 images a day. The image is shrunk in your browser, processed in memory and never stored by us; press Edit in the studio to tweak the result.