A Gemini AI photo prompt works best as three plain sentences: what to keep, the one thing to change, and the look you want. Below are 48 original prompts for portrait retouching, outfit and scene swaps, product shots, consistent characters, restoration and styles, each with a tip. Pick Flash or Pro in the Gemini app for photo edits, expect a SynthID watermark on every result, and only edit people who agreed to it.
A Gemini AI photo prompt is a plain-language instruction you type next to an uploaded photo, or on its own, in the Gemini app, and the ones that work name what to keep, the one thing to change and the look of the result. Below are 48 original, copy-ready prompts grouped by job, each with what it does and a tip, plus the model settings, limits and watermark rules that decide what you get back.
Checked October 2026 against Google's own pages: Generate and edit images with Gemini Apps, Gemini Apps limits, the Gemini image generation page, the Google AI plans page, verifying AI-generated images, SynthID Detector and the Generative AI Prohibited Use Policy. Google changes model names often; the settings below are the ones its pages showed on 16 October 2026.
The prompts are written for your own photos, photos of people who said yes, products you sell and fictional characters. If you are building a recurring AI character, start with our consistent character guide, which compares every method, then come back for the scene prompts.
How to use these prompts in the Gemini app today
Open gemini.google.com or the Gemini app, sign in, attach your photo and paste a prompt. Google's help centre lists three ways to edit: change an image Gemini made, upload a photo and ask for edits, or upload several photos and combine them into a new one.
- 1Sign in and pick the model
Choose Flash or Pro in the model menu if your plan has them. They run Nano Banana 2, which handles uploads and follow-up edits.
- 2Attach the photo
Use the plus or add-files button to upload from your device. For a combine prompt, upload the person first and the garment or product second.
- 3Paste one prompt
Keep it to one change. Expected: a single edited image in the reply.
- 4Refine in the same chat
Ask for the next change as a follow-up so Gemini works from its last result, not from scratch.
- 5Redo with Pro if you pay
On a Google AI plan, the More menu offers Redo with Pro, which regenerates with Nano Banana Pro.
- 6Download and check
Download is 1K on free and 2K on a paid plan. Zoom in on faces, hands and any text before you post.
Source: Gemini Apps Help, checked October 2026
Which image model answers depends on the model setting, and Google's pages do not line up perfectly. The help centre pairs each setting with an image model, the plans page lists Flash-Lite as the free model, and the image page calls Nano Banana 2 free and names Nano Banana 2.1 as the best model for paid plans. In practice: if your menu offers Flash or Pro, use it for anything with an uploaded photo.
| Model setting | Image model (help centre) | Good for | Who gets it |
|---|---|---|---|
| Flash-Lite | Nano Banana 2 Lite | Fast single generations. Google says it is not designed for multiple image references or multiple edits | The model the plans page lists for free accounts |
| Flash or Pro | Nano Banana 2 | Edits of an uploaded photo, several reference photos at once, follow-up edits | Flash from Google AI Plus, Pro from Google AI Pro |
| Redo with Pro | Nano Banana Pro | Regenerating a finished image at higher quality from the More menu | Google AI plans, and not after the day's Nano Banana 2 quota is used up |
The limits that matter for photo work, as Google lists them (checked October 2026):
| What | Rule | Source |
|---|---|---|
| Who can edit photos | Signed-in personal accounts aged 18 or over; generating alone needs 13 or over | Gemini help centre |
| Download size | 1K without a Google AI plan, 2K with one | Gemini help centre |
| Usage limits | Standard on free; 2x on Google AI Plus; 4x on Google AI Pro; Ultra 5x or 20x the Pro limit | Gemini limits page |
| When limits reset | Every 5 hours, until you reach the weekly limit | Gemini limits page |
| Image quota | Google publishes no number of images per day for any plan | Gemini limits page |
| Watermark | Invisible SynthID watermark plus a visible watermark on generated images | Gemini image page |
Sources: Gemini help centre, limits page, image page and plans page. On 16 October 2026 the plans page showed Google AI Ultra at $99.99 a month (5x the Pro limits) or $199.99 (20x) in the United States. Google says media generation uses more of your limit than text.
The Gemini AI photo prompt formula: keep, change, look
Every prompt below follows the same order, because it is the order an editor would brief a retoucher in. Google's own advice is to start from subject, action and scene, then add composition, style, quality and aspect ratio. For edits, put the protection first.
- 01Keep
What must not move: face, expression, hair, pose, background, label text.
- 02Change
The one thing to edit, described by material, colour and fit, not by a vague adjective.
- 03Light
Where the light comes from and what kind: window, low sun, single key light.
- 04Format
Aspect ratio as numbers, crop, and anything to leave out, such as borders or date stamps.
- "Make me look better"
- "Change outfit and background and make it sunset"
- "Professional headshot"
- No mention of the face
- "Enhance" or "retouch" with no target
- "Keep my face, hair and pose exactly as they are"
- One change per message, the next one as a follow-up
- "Single soft key light from the right, charcoal background"
- The fixed parts named before the change
- The exact fix: "soften the redness on my cheeks"
Operator practice from editing creator photos, not a Google rule.
48 Gemini AI photo prompts, grouped by use
Copy a prompt, swap the details (garment, scene, colours) for your own, and send it with your photo attached. All 48 were written for this page.
Portrait retouch prompts
Upload your own portrait first. These prompts all start by fixing the face, because an unconstrained retouch is where Gemini drifts.
Keep my face, expression, hair and pose exactly as they are. Even out the skin tone and soften the redness on my cheeks, but leave pores, freckles and fine lines visible. No smoothing filter look.
What it does: A light, believable skin pass. Tip: If it still looks plastic, add "match the skin texture of the original photo at 100% zoom".
Relight this photo as if I were sitting next to a large north-facing window on an overcast day. Keep everything else the same: face, clothes, background, framing.
What it does: Swaps harsh flash or overhead light for soft window light. Tip: Name a real light source. "Better lighting" gives Gemini nothing to aim at.
Remove the glare on my glasses lenses so my eyes are clearly visible. Do not change the frames, my eyes or anything else in the image.
What it does: Cleans lens reflections. Tip: Works best when the eyes are at least partly visible behind the glare.
Tidy the flyaway hairs around my head and shoulders against the background. Keep my hairstyle, colour and length unchanged.
What it does: The edit retouchers charge for, done in one line. Tip: Add "especially on the left side" if one area matters most.
Change the background behind me to a plain warm grey studio backdrop with a soft gradient. Keep my face, clothes, pose and the light on my face unchanged.
What it does: A clean profile or speaker photo from a busy shot. Tip: Mention the light on your face, or Gemini may relight you to match the new backdrop.
Whiten my teeth slightly, to a natural off-white. Keep the shape of my teeth, my smile and my lips exactly as they are.
What it does: A subtle dental touch-up. Tip: "Slightly" and "natural off-white" keep it from turning into bright-white veneers.
Crop and extend this photo to a 4:5 portrait frame for Instagram, with my face in the upper third. Extend the background naturally where needed. Do not change me.
What it does: Reframes a landscape shot for a feed post. Tip: Give the ratio as numbers; Google's own example prompts do the same.
Remove the person in the background on the right and fill the space with the same street scene. Leave me and the rest of the photo untouched.
What it does: Removes a photobomber. Tip: Point to the object by position ("on the right", "behind my shoulder"), not just by name.
Outfit and scene swap prompts
These change one large thing and protect the rest. Keep each request to one change; stack the next one in a follow-up message.
Keep my face, hair, body and pose exactly the same. Replace my T-shirt with a navy linen shirt, sleeves rolled to the elbow, top button open. Match the lighting of the original photo.
What it does: A single garment swap. Tip: Describe fabric and fit, not just colour. "Navy shirt" leaves Gemini to invent the rest.
Using the second image as the outfit, dress the person in the first image in that exact jacket, keeping its colour, buttons and pattern. Keep the first image's face, pose and background.
What it does: Puts a real garment from a product photo onto your photo. Tip: Upload the person first and the garment second, and say which is which.
Move me from this living room to a quiet cafe terrace in late afternoon sun. Keep my face, outfit and pose. The light on me should come from the low sun on the left.
What it does: A full scene swap with matched light. Tip: Telling Gemini where the new light comes from is what stops the cut-out look.
Change the season in this photo from summer to a light snowfall in winter. Add a wool scarf to my outfit. Keep my face and pose unchanged.
What it does: Seasonal variant of an existing shot. Tip: Add one matching wardrobe detail so the person does not look pasted into the weather.
Keep everything the same but change the time of day to blue hour just after sunset, with warm window lights on in the buildings behind me.
What it does: A mood change without moving location. Tip: Good for testing which version of a photo reads better before you reshoot.
Swap my sneakers for brown leather loafers with no socks. Keep my trousers, pose and the floor exactly as they are, and keep the shadow under my feet.
What it does: Footwear swap, the hardest small edit. Tip: Mentioning the shadow stops the floating-feet effect.
Put me in a minimalist office with white walls, a wooden desk and one plant, as if photographed with a 35mm lens at eye level. Keep my face, hair and clothes unchanged.
What it does: A work-profile setting from a casual photo. Tip: A lens and camera height make the space feel photographed rather than generated.
Make the outfit in this photo all black: black shirt, black trousers, black boots. Keep the cut and fit of every garment the same as now.
What it does: A colour-only restyle. Tip: "Keep the cut and fit" is what separates a recolour from a new outfit.
Product shot prompts
Upload a clear photo of the real product. Gemini is good at staging and poor at keeping small print exact, so check every label before you sell with the image.
Place this product on a seamless pure white background with a soft shadow directly beneath it, front view, square 1:1 frame. Keep the label, logo and colours exactly as in my photo.
What it does: A marketplace-style main image. Tip: Read the label in the result letter by letter. Small text is where errors appear.
Stage this candle on a light oak side table next to a folded linen throw and a short stack of books, warm evening lamp light from the left, shallow depth of field.
What it does: A lifestyle shot from a plain product photo. Tip: Three props is plenty. Long prop lists crowd the product out of focus.
Show this skincare bottle on wet grey stone with a few water droplets on the bottle, cool morning light, close-up at a slight downward angle. Do not change the bottle shape or text.
What it does: A textured editorial look. Tip: Name the surface material; it sets the whole mood.
Create a flat lay seen from directly above: this notebook in the centre, a pen, a coffee cup and a phone arranged around it on a sand-coloured desk, even daylight, 4:5 frame.
What it does: A top-down flat lay. Tip: "Seen from directly above" is clearer than "flat lay" alone.
Make three versions of this product photo, each on a different solid background: sage green, terracotta and pale blue. Keep the product, angle and lighting identical across all three.
What it does: Colour tests for ads or a storefront. Tip: Ask for all three in one message so the product stays consistent between them.
Show this mug being held by two hands in a knitted sweater, cropped so no face is visible, cosy kitchen background softly out of focus.
What it does: A human-scale shot without a model. Tip: Cropping out the face keeps the image about the product and avoids an invented person.
Turn this phone photo of my product into a clean e-commerce image: straighten it, remove the clutter behind it, fix the white balance so the white packaging reads as white.
What it does: A rescue edit for a rushed photo. Tip: List the fixes in the order you would do them by hand.
Place this backpack on a mountain trail at sunrise, resting against a rock, with mist in the valley behind. Keep the backpack's colours, zips and logo exactly as photographed.
What it does: An outdoor scene for a product that never left the studio. Tip: If you sell it, label staged scenes honestly; see the ecommerce guide linked below.
Staged scenes of a real product are fine as long as the product itself is shown truthfully. Our guide to AI model photos for ecommerce covers the fidelity check and what each sales channel wants labelled.
Consistent character prompts
For a recurring fictional character, use one fixed character line and change only the scene. These six are starters; the full set of 30 is in our face consistency pack.
This is my character, Mara: the woman in the uploaded image. In every image I ask for, keep her face, eye colour, hairline and the small scar on her chin identical. First scene: reading on a train, window light.
What it does: Opens a chat with a fixed identity. Tip: Give the character a name and one distinctive detail you can check at a glance.
Same character as before. New scene only: she is ordering at a street food stall at night, neon light from the stall. Same face, hair and build.
What it does: A follow-up scene in the same chat. Tip: Say "same character as before" every turn; it reminds the model what is fixed.
Show the same character from the side, three-quarter profile, looking out to sea on a windy beach. Keep the face shape and nose profile consistent with the front view.
What it does: A new angle of a known face. Tip: Angle changes drift the most. Generate a few and keep the closest one.
Using images one and two as the same character, create a photo of her laughing with a friend at a picnic. The friend is a different person; only my character must match.
What it does: A two-person scene with one fixed face. Tip: Say who must match and who is free, or Gemini may blend the two faces.
Same character, same outfit as the last image. Change only her expression to a surprised half-smile. Nothing else moves.
What it does: Expression change without a new image. Tip: Small, single changes keep identity far better than a brand-new scene.
Create a 2x2 grid of the same character in four outfits: running gear, a suit, a raincoat and a summer dress. Same face and hair in all four panels.
What it does: A wardrobe sheet for planning a feed. Tip: Check each panel separately; a grid can hide one panel that drifted.
Gemini holds a face well across a few scenes, not across a whole feed. For 30 more templates with a drift-risk rating, use our Gemini face consistency prompt pack. Running a character as a business, with a trained face model, video and a posting plan, is what the AI Influencers program covers, from persona creation and LoRA training to photorealism and AI video.
Photo restoration prompts
Scan or photograph the print flat and in even light first. Restoration prompts should repair, not reinvent, so say so in every one.
Restore this old photo: remove the scratches, dust spots and the crease across the top. Do not change anyone's face, expression or clothing. Keep it black and white.
What it does: A conservative clean-up. Tip: Start here before colourising. Fix damage first, then add colour in a second turn.
Colourise this black-and-white photo with natural, slightly muted colours typical of a 1960s print. Keep faces, clothing details and the grain of the original.
What it does: Period-appropriate colourisation. Tip: Naming the decade steers the palette away from modern saturated colour.
The bottom-left corner of this photo is torn off. Rebuild only that corner so it matches the surrounding background. Do not change the people.
What it does: Fills a missing piece. Tip: Gemini invents what was in the missing area, so do not treat it as a record.
This photo is faded and yellowed. Restore the contrast and correct the colour cast so whites look white, without sharpening faces or adding detail that is not there.
What it does: Fixes fading and colour cast. Tip: "Without adding detail that is not there" limits invented skin and eye detail.
Sharpen this blurry family photo just enough that faces are recognisable, but do not change their features, age or expressions. If a face is too blurry to recover, leave it soft.
What it does: Careful de-blurring. Tip: Compare with the original. If a relative no longer looks like themselves, go back a step.
Remove the shop stamp and handwritten date from the lower edge of this print, and fill the area to match the photo. Keep everything else exactly as it is.
What it does: Removes marks from the print border. Tip: Keep an untouched scan of the original; the stamp and date may matter to your family history.
Straighten this photo, which I took at an angle of a framed print, and remove the reflection from the glass. Keep the photo inside the frame unchanged.
What it does: Fixes a phone snapshot of a framed picture. Tip: Better still, take the print out of the frame and photograph it flat.
Restore this photo and return it at the same size and framing as the original, with no added border, caption or vignette.
What it does: Stops unwanted extras. Tip: Add this line to any restoration prompt that came back with decoration you did not ask for.
Style and creative prompts
These work on your own photos or from text alone. Describe the medium and the light, and name no living artist; describe the look instead.
Turn this photo of me into a watercolour portrait on textured cotton paper, loose brushwork, the background left mostly white. Keep my likeness recognisable.
What it does: A painted portrait from a photo. Tip: Naming the paper and the brushwork beats "make it artistic".
Recreate this photo as if shot on 35mm film in the 1990s: soft grain, slightly warm colours, a little lens flare, date-stamp free.
What it does: A film look. Tip: "Date-stamp free" stops the orange date that the model loves to add.
Make this photo look like a still from a moody detective film: low-key lighting, deep shadows, cool blue-green tones, wide 21:9 frame.
What it does: A cinematic grade and crop. Tip: Give the frame ratio; it changes the composition, not just the crop.
Convert this street photo into a clean flat vector illustration with five colours, bold shapes and no outlines, suitable for a poster.
What it does: A graphic poster version. Tip: A colour count keeps the result simple enough to print.
Create a high-end editorial portrait from this selfie: medium shot, single soft key light from the right, dark charcoal background, natural skin, sharp eyes.
What it does: A studio-style portrait from a phone selfie. Tip: Start from a sharp, well-lit selfie; Gemini improves light more reliably than focus.
Make a sticker of my dog from this photo: cut out, thick white border, slightly cartoon style but clearly the same dog, transparent-looking plain background.
What it does: A pet sticker. Tip: Gemini returns an image on a background; remove it in an editor if you need true transparency.
Turn this room photo into a pencil architectural sketch with light grey shading, keeping the furniture layout accurate.
What it does: A hand-drawn version of a real space. Tip: Useful for mood boards and listing ideas; keep the real photo for anything factual.
Create a photo of a ceramic coffee cup on a balcony railing at sunrise over a city, steam rising, shot at eye level with an 85mm lens, shallow depth of field, 3:2 frame.
What it does: A text-only scene with photographic language. Tip: Subject, action, scene, then camera: Google's own prompt advice in that order.
Make a clay stop-motion version of this family photo: everyone as handmade plasticine figures, visible fingerprints in the clay, soft studio light.
What it does: A playful 3D style. Tip: Texture details ("visible fingerprints") are what sell a material style.
Turn this landscape photo into a vintage travel poster with a big title at the top that reads "LAKE BLED", flat colours and a 2:3 frame.
What it does: A poster with text. Tip: Put the exact words in quotes and keep them short; check the spelling in the result.
Gemini photos and the SynthID watermark
Every image Gemini generates is marked. Google's image page says it adds an invisible SynthID watermark and a visible watermark to show the image is AI-generated, and that you can upload an image to Gemini and ask whether Google AI made it.
Disclosure rules depend on where you post. Instagram's rules are in our guide to Instagram AI labels. A light retouch of your own photo and a fully generated scene are treated differently on most platforms, so read the rule for the edit you made.
Consent and likeness: never edit a real person without a yes
The rule for this page is simple: edit your own photos, photos of people who agreed to the edit, or characters you invented. Google's Generative AI Prohibited Use Policy bans content that violates others' privacy, impersonating an individual without disclosure in order to deceive, and non-consensual intimate imagery, and Google says Gemini may remove images made from prompts that may break it.
- It is you, or the person in the photo said yes to this specific edit
- If a child is in the photo, keep the edit to restoring or cleaning up your own family picture
- The edit does not put words, acts or settings on a real person they never had
- No celebrity, public figure or ex-partner is the subject of a swap or a scene
- Product and label details still match the real product
- You know whether the platform you post on wants an AI label for this kind of edit
- An untouched original is saved, in case anyone asks what was changed
Swapping one person's face onto another's photo is its own legal area. Our AI face swap guide covers consent and the law, and why a persona usually needs a different tool.
Why a Gemini photo edit goes wrong, and the fix
Most failed edits come from the prompt, not the model. Here are six common problems, with the line that fixes each.
| Problem | Usual cause | Fix |
|---|---|---|
| The face changed | The prompt never said what to keep, or asked for several changes at once | Add "keep my face, hair and pose exactly as they are" and change one thing per message |
| It looks pasted in | The new scene's light does not match the person | Name the light source and direction: "low sun from the left" |
| Skin looks plastic | "Retouch" or "enhance" read as a smoothing filter | Ask to keep pores and texture, and name only the specific fix |
| Label text is wrong | Small print is regenerated, not copied | Ask to keep the label exactly, then check it letter by letter or fix it in an editor |
| Edits stop working after a few turns | Flash-Lite is not designed for multiple edits | Switch to Flash or Pro, or start a new chat from the best image so far |
| It refused the request | The prompt may have touched a policy area, such as editing a real person | Use your own photo or a fictional character; see the consent section |
Gemini AI photo prompts: FAQ
What is a good Gemini AI photo prompt?
A good Gemini AI photo prompt says three things in plain sentences: what to keep from the uploaded photo, the one thing to change, and the look of the result, such as lighting, lens feel and aspect ratio. "Keep her face, pose and the background exactly as they are. Change only the jacket to a camel wool coat. Soft daylight, natural skin texture" beats a list of keywords.
Is Gemini AI photo editing free?
Yes, with limits. Google lists image generation and editing in the free Gemini plan, and its image page calls Nano Banana 2 free of charge. Free accounts download at 1K and paid plans at 2K, per the Gemini help centre, and paid plans get higher usage limits. You must be signed in, and editing photos needs an account holder aged 18 or over. Checked October 2026.
Which Gemini model should I pick for photo edits?
Pick Flash or Pro in the model menu if your plan offers them. Google's help centre says those settings run Nano Banana 2, which accepts multiple reference images, while Flash-Lite runs Nano Banana 2 Lite, which is built for speed and not designed for multiple references or repeated edits. Paid subscribers can also redo an image with Nano Banana Pro. Checked October 2026.
Do Gemini photos have a watermark?
Yes. Google's image page says Gemini adds an invisible SynthID watermark and a visible watermark to images it generates. The SynthID mark sits in the pixels and is built to survive resizing, recolouring and compression, so cropping out the visible mark does not make an image unmarked. Anyone can check a file in the Gemini app or at synthid.com. Checked October 2026.
Can I use Gemini to edit a photo of another person?
Only with that person's consent, and never to deceive. Google's generative AI use policy bans violating others' privacy, impersonating an individual without disclosure in order to deceive, and non-consensual intimate imagery. Beyond the policy, many places now have deepfake and likeness laws. Edit your own photos, photos of people who agreed, or fictional characters you created.
Why does Gemini change the face when I edit a photo?
Usually because the prompt asked for too much at once or never said what to keep. Name the face, hair and pose as fixed, change one thing per turn, and keep follow-up edits in the same chat so the model works from its last result. Flash-Lite also struggles more with repeated edits, so switch to Flash or Pro for a long editing session.