Gemini AI Photo Prompts: 18 to Edit, Restore, Restyle

In brief

A good Gemini AI photo prompt names one edit and lists what must stay the same, and this page gives 18 to paste. As of November 2026 Gemini edits uploaded photos on every plan with a Nano Banana model, Google prints no daily image cap, and editing needs an account aged 18 or over.

What makes a good Gemini AI photo prompt?

A good Gemini photo prompt asks for one change and lists what must stay untouched. Google's own prompting tips say the same thing: "For modifying an existing image, be direct and specific." Its example is "change the man's tie to green, remove the car in the background."

A stack of blank instant photo prints on a charcoal plinth with the rainbow Gemini sparkle standing on top, a vintage camera on the left plinth and a terracotta banana on the right plinth, in a dark studio under white rim light.

Use this three part shape for every prompt below: the edit, the keep list, the look. Here is one filled in. It needs a selfie and runs on every plan:

"Change my jacket to a navy blazer over a white shirt. Keep my face, hair, expression and pose exactly as they are. Soft window light from the left, plain grey wall behind me."

The keep list does the most work. Without it, an image model repaints the whole frame, and faces drift first. The 18 prompts below are ours, written to that shape. Google publishes only a handful of edit prompts, so we haven't copied any from its pages. Competing lists run from 18 to 51 prompts by their titles. We stopped at 18 because each one needs its own keep list, and trend prompts would only add near duplicates.

Which Gemini plan and model runs these photo prompts?

Every Gemini account can edit an uploaded photo with Nano Banana 2, and paid plans add a Pro redo and bigger downloads. Google's image help page says Nano Banana 2 is used when the Gemini model is set to Flash or Pro, and that it "can accept multiple reference images." We read the page and the limits page on 15 November 2026:

PlanImage modelDownload sizeUsage vs standard limitsRedo with Nano Banana Pro
Without a Google AI planNano Banana 2 (Nano Banana 2 Lite on Flash-Lite)1KStandardNo
Google AI PlusNano Banana 2 (overview page: 2.1)2K2xYes
Google AI ProNano Banana 2 (overview page: 2.1)2K4xYes
Google AI UltraNano Banana 2 (overview page: 2.1)2K5x or 20x AI ProYes

Google prints no number of images per day on either page, only checkmarks, so any "20 a day" or "100 a day" you read now describes an older version of the page. Gemini image generation limit has the last printed caps and the date they vanished. Two rules from the help page matter for editing. Every limit refreshes "every 5 hours until you reach your weekly limit." And a Pro redo needs quota left: "If you reach your daily quota of Nano Banana 2 images, you can't redo any additional images with Nano Banana Pro."

Watch the model picker. The help page says Nano Banana 2 Lite is used on Flash-Lite and "isn't designed for multiple image references or for multiple edits." So pick Flash or Pro before any prompt that uses two photos or stacks several changes. One more oddity: Google's image generation overview names Nano Banana 2.1 as the model for AI Plus, Pro and Ultra, while the help page still says Nano Banana 2. Google's two pages disagree as of today.

How do you edit an uploaded photo in Gemini, step by step?

You edit a photo by uploading it to a Gemini chat and typing the change: open gemini.google.com, upload the photo, send one edit with a keep list, then fix it with follow-ups. Two details from Google's pages save a failed first try.

  1. Age and account. Editing is for personal accounts aged 18 or over. A 13 year old can generate images but not edit them.
  2. Reference photos. Google's October 2025 tips post says you can combine up to three images into one. The developer image docs say Nano Banana models allow you to mix up to 14 reference images through the API.

Stay in one chat while you refine. A reply like "make the sky less orange, change nothing else" edits the latest version, so you spend one more small edit instead of a fresh start. Four small edits are fewer image requests than eight full retries.

Which Gemini prompts fix and restyle a portrait?

Four prompts cover most portrait jobs: a headshot, a lighting fix, a hairstyle try-on and a lifestyle scene. Upload a clear photo of yourself first. Each keeps the face fixed, because the face is the part you care about.

  1. Headshot. "Turn this selfie into a head and shoulders headshot. Keep my face, hair and expression exactly as they are. Dress me in a charcoal blazer over a white shirt, with a soft grey wall behind me and gentle window light from the left."
  2. Lighting fix. "Relight this portrait with warm sun from a window on the right, as if it were an hour before sunset. Lift the shadows under my eyes a little. Keep my skin texture natural and change nothing else."
  3. Hairstyle and glasses. "Show me with a shorter, textured haircut and thin round glasses. Keep my face shape, skin tone, clothes and background the same."
  4. Lifestyle scene. "Place me on a quiet city street at dusk with shallow depth of field and lit shop windows behind. Keep my face, outfit and pose from the original photo."

Google's prompting post for Nano Banana Pro cautions that "character consistency across edits may vary." Compare the result to your original before you use it as a profile photo.

Which Gemini prompts make good product photos?

Four prompts cover the usual shop needs: a clean white shot, a lifestyle scene, a catalog cutout and a desk scene. Upload a plain photo of the product on any surface. The keep list protects the label, which is what a model changes first.

  1. White acrylic shot. "Place this product on a white acrylic table with a soft shadow underneath. Keep the label sharp and readable and don't change the product's shape or colors. Square crop, even studio light."
  2. Outdoor scene. "Set this bottle on a mossy stone beside a stream in morning mist, shallow depth of field. Keep the bottle and its label exactly as photographed."
  3. Catalog cutout. "Cut this product out and put it on a flat terracotta background with one hard shadow to the right, like a catalog page. Keep the product itself unchanged."
  4. Desk scene. "Show this mug on a wooden desk beside a laptop and a notebook in warm morning light. Same mug, same handle angle, same print on the side."

Google's developer docs show a product prompt of the same kind: a "studio-lit product photograph" with sharp focus on a detail. Ask for one shot per message.

Can Gemini restore an old photo, and which prompt works?

Yes, Gemini can clean and colorize an old photo, but it repaints pixels and can change faces. Google's Nano Banana tips post lists restoration among the uses, without a prompt. Scan the print at a high resolution first, then try these three in order:

  1. Clean up. "Restore this scanned family photo. Remove scratches, creases and dust and sharpen the faces a little. Keep every face, outfit and pose as in the original, and keep it black and white."
  2. Colorize. "Colorize this black and white photo with muted, natural colors that suit the era. Keep faces and skin tones realistic and don't add any objects."
  3. Repair damage. "Repair the torn corner and the water stain on the left by continuing the wallpaper and floor pattern. Don't touch the people."

Hold your original next to the result and check the eyes, the teeth and any hands. If a face has drifted, send "go back to the original face and keep the rest of this restoration" instead of rerunning from scratch. Never delete the scan.

Which prompts change the style or scene of a photo?

Seven prompts change a photo without changing the people in it: three style swaps, a cleanup, a sky swap, a sign fix and a two photo blend. Google's Nano Banana Pro tips, dated 20 November 2025, show the same plain style of request, such as "Turn this scene into nighttime."

  1. Watercolor. "Turn this photo into a watercolor painting with loose brush edges and visible paper texture. Keep the composition and the poses."
  2. Flat illustration. "Redraw this photo as a flat vector illustration using five colors and bold outlines. Keep the layout the same."
  3. Film look. "Make this photo look like 35mm film from 1994: visible grain, slightly faded colors, soft direct flash."
  4. Remove clutter. "Remove the trash can and the parked car from the background and fill the gaps with matching pavement and wall. Leave the people untouched."
  5. New sky. "Change the sky to a clear sunset with a few thin clouds, and match the light on the subject to it."
  6. Fix a sign. "Replace the text on the shop sign with the words OPEN LATE. Keep the sign's font, color and angle."
  7. Blend two photos. "Put the dog from the second photo on the sofa in the first one. Match the light and scale, and keep the room unchanged."

Prompt 17 depends on text rendering. Google's developer docs list "Advanced text rendering," yet its Nano Banana Pro prompting post warns that "producing accurate spellings may not work perfectly." Check every letter. Prompt 18 needs the Flash or Pro setting, not Flash-Lite. We haven't benchmarked these 18 prompts against each other, so treat them as starting points and adjust the keep list to your photo.

What if Gemini stops mid-edit, and where else can you run these?

If Gemini hits your limit, read the refresh time in its notice and draft the next prompt in the meantime. Google says that when you reach your limit "Gemini Apps gives you another notification that tells you when your limit will refresh." The same prompts work in any app that edits an attached photo, and ChatGPT is the other big one: can ChatGPT edit photos covers what OpenAI's plans allow.

Whizi, which publishes this page, has a Nano Banana row in its picker for editing a photo you attach, with up to 4 source photos per edit. The free plan lets you upload 5 photos a day and make 2 pictures a month, and Pro includes 100 a month (how images work in Whizi). It loses to Gemini in three ways. Whizi runs Nano Banana 2 Lite, which Google calls not designed for multiple edits, so stack fewer changes per prompt. There is no Pro redo. And Google prints no cap for Gemini's free plan, while Whizi prints a small one.

What does Gemini refuse to edit, and what does it add to every image?

Gemini refuses edits that break Google's Generative AI Prohibited Use Policy, and it marks every image it makes or edits. Google's policy, last modified 17 December 2024, bars content that "Facilitates non-consensual intimate imagery," content "Relates to child sexual abuse or exploitation," and "Impersonating an individual (living or dead) without explicit disclosure, in order to deceive." It also bars using "personal data or biometrics without legally-required consent." Google's image page adds that Gemini may remove an image when its systems detect a possible violation.

Google publishes no list of named people Gemini will not edit, so we can't give you one. In practice, a photo of someone else is the riskiest upload: their face is personal data, and an edit meant to deceive is covered by the policy. Rewording a refused prompt to slip past the filter is itself listed: "Circumvention of abuse protections or safety filters." Edit photos of yourself, or of people who said yes.

Every result carries marks, per Google's watermark help page: "All AI-generated media created or edited with Gemini Apps includes an invisible SynthID watermark," plus Content Credentials (C2PA) metadata. A visible watermark is a separate setting under Settings, Media Watermark, and it is only offered to AI Ultra subscribers in India, South Korea and Vietnam. Switching it off "doesn't affect SynthID watermarks or Content Credentials."

You can test an image yourself. Google's verification page says to upload one file of 100 MB or less and ask whether Google AI made it, with a quota of "approximately 10 image checks" in any 24-hour rolling window. It also warns that Gemini "can currently only recognize content created by Google AI tools," and that a tiny edit may carry no detectable watermark.

Workflow checklist
  • Name one edit per prompt, then list what must stay the same
  • Pick the Flash or Pro setting, not Flash-Lite, for multi photo or multi edit prompts
  • Stay in one chat and send small follow-ups instead of rerunning
  • Compare restored faces to the original scan before you keep a result
  • Edit photos of yourself or of people who agreed
  • Read the refresh time in the limit notice before you retry
Common questions

Frequently asked questions

Is Gemini photo editing free?

Yes, on a free account with limits, as of November 2026. Google's limits page shows Nano Banana 2 image generation with a checkmark for accounts without a Google AI plan, and downloads are 1K instead of 2K. Google prints no daily image count, so the notice in the app tells you when your limit refreshes.

Can Gemini restore and colorize old photos?

Yes. Google's Nano Banana tips post lists restoring and colorizing among the uses of its editing model. Upload a scan, ask for dust and scratch removal with a keep list for faces and clothes, then check the result against the original, because the model repaints the image and can alter faces.

Does Gemini add a watermark to edited photos?

Yes. Every image Gemini creates or edits carries an invisible SynthID watermark and Content Credentials, and neither can be switched off. The visible mark is a separate setting under Settings, Media Watermark, and in India, South Korea and Vietnam only AI Ultra subscribers see that setting.

Can I paste someone else's photo into a Gemini prompt?

You can upload it, but Google's Prohibited Use Policy bars edits that use personal data or biometrics without legally required consent or that impersonate a person to deceive. Google lists no names, so Gemini decides case by case and may remove the image. Stick to your own photos or ones the person agreed to.