Gemini AI Photo Editing Prompts to Copy and Paste

To edit a photo in Gemini, upload it first, then paste a prompt that says exactly what changes and, just as clearly, what must stay the same, because that last line is what keeps your face and the rest of the picture intact. Gemini's image model (Google's developer docs call it Nano Banana) responds best to prompts built from four parts: the subject, the exact change, the style reference, and what must stay untouched. Below are ready-to-use prompts for portraits, backgrounds, outfits, lighting, old photo restoration and group shots, plus how to check faces and skin tones before an edit goes anywhere public.

Why most copied prompts come back looking wrong

If you've pasted a prompt from a random listicle and got a melted face or a hand with seven fingers, the wording was usually missing one part: the line that says what stays the same. Gemini's image model isn't reading your prompt like a creative brief, it's reading it like an instruction set, and vague instructions get vague results.

A prompt that works reliably has four parts in this order: who or what is in the photo, exactly what changes, what style or reference it should match, and what must stay untouched. Drop any one of those four and the model fills the gap with a guess, which is usually where things go wrong. "Make this photo look professional" is a wish. "Keep the person's face, pose and clothing exactly as they are, replace the cluttered kitchen background with a plain soft grey studio backdrop, and add even studio lighting from the left" is an instruction.

Google uses the same idea in its own image generation docs for developers. Its editing template reads: "Using the provided image, change only the [specific element] to [new element/description]. Keep everything else in the image exactly the same, preserving the original style, lighting, and composition." Filled in, it becomes: "Using the provided image, change only the shirt to a navy linen shirt. Keep everything else in the image exactly the same, preserving the original style, lighting, and composition." When you're unsure how to phrase an edit, paste that pattern and swap only the bracketed parts.

A worked example: turning a phone photo into a usable listing photo

Say you run a small home bakery and you've taken a photo of a cake on your kitchen counter, with a mug, some crumbs, and your washing basket visible in the background. You want it to look like something you'd see on a bakery's Instagram grid, not a snapshot.

Instead of asking Gemini to "make it look nicer," try this exact prompt:

• "Keep the cake exactly as it is, including its colour, texture and the plate it's sitting on. Remove everything else from the background and replace it with a clean white marble surface and soft natural window light coming from the left. No text, no props, no hands in frame."

Notice what's happening: the cake (subject) is protected, the background swap is specific (white marble, not "a nice background"), the lighting direction is named, and there's an explicit exclusion list at the end. That exclusion list is doing more work than people realise, because it stops the model adding a stray spoon or a logo that was never there.

If the first result still has a shadow in the wrong place, don't start over. Add one line to the same conversation: "Keep everything the same, just move the light source slightly more overhead so the shadow falls directly behind the cake." Google's docs describe Gemini's image generation as something you create, edit and iterate on conversationally, so refining in small steps usually gets you closer than rewriting the whole prompt. You'll notice most of the prompts on this page open by naming what to keep. That's deliberate: say what must stay the same before you say what should change, so the edit has a fence around it and your face, background or product isn't up for grabs.

Here's how a typical run goes, one change per message:

• Result 1 has a harsh shadow on the left of the cake. Send: "Keep everything the same, just soften the shadow."
• Result 2 has added a spoon you never asked for. Send: "Remove the spoon, nothing else changes."
• Result 3 has garbled the lettering on the cake box. Send: "Leave all packaging text exactly as in the original, or blur it."

Three short messages, three fixes, and the cake itself never moved. Bundle all three into one message and you risk trading one problem for another.

Can ChatGPT make a professional headshot? These 20 edits show what it can do with your own photo.

I've written up the fix for the ChatGPT face change problem when you edit photos.

Portrait and personal photo prompts

These are the ones most people want, whether it's a LinkedIn headshot, a dating profile photo, or something just for fun.

• "Keep my face, expression and hairstyle exactly as they are in this photo. Change only the outfit to a tailored navy blazer over a white shirt, and replace the background with a softly blurred neutral office setting."
• "Keep my face and pose unchanged. Adjust the lighting to warm golden-hour light coming from behind and slightly to the right, and add a subtle lens flare. Do not alter my skin tone or facial features."
• "Turn this photo into a professional studio headshot. Keep my face identical, add a plain charcoal grey backdrop, soft even lighting from both sides, and a slight depth-of-field blur behind me."

If the face changes anyway, send this in the same chat: "The face has changed. Restore the face from the original upload exactly; change only the background." Still off? Start a new chat, upload the original again and use Google's "change only the [element]" wording from above. Naming the features to protect works better than "my face": "keep my hairline, glasses and beard identical" gives the model something concrete to hold on to.

If you want something more playful than professional, it's worth looking at a dedicated set of funny AI prompts for pictures of yourself, because the prompt structure for comedy (exaggeration, absurd setting, specific prop) is almost the opposite of the structure for a polished headshot.

Background and setting swap prompts

• "Keep the subject, their pose and clothing exactly as they are. Replace the background with a quiet library interior, warm wooden shelves, soft diffused light, nothing else changed."
• "Keep the subject unchanged. Replace the background with a blurred city street at night, with soft bokeh lights, as if shot on a wide aperture lens."
• "Keep the subject in the same position and lighting direction. Replace only the background with a plain cream studio backdrop suitable for an e-commerce product listing."

Outfit, colour and detail changes

• "Keep the person's face, pose and background exactly the same. Change the outfit to a red knee-length dress with a fitted waist, and adjust shadows on the fabric to match the existing light source."
• "Keep everything in the photo unchanged except the colour of the car, which should become matte dark green instead of silver. Match the reflections to the new colour."
• "Keep the composition identical. Remove the person in the background on the left, and fill the space naturally with the existing wall and floor pattern."

Restoring and colourising old photos

This is one of the places Gemini earns its keep, because it's doing something a basic filter can't: inferring plausible colour and texture from a black and white source.

• "Colourise this black and white photo with natural, realistic skin tones and period-accurate clothing colours for the 1950s. Keep the composition and facial details exactly as they are, do not add or remove anything."
• "Restore this damaged photo. Remove scratches, creases and discolouration, sharpen the facial details, and keep the original expression and pose unchanged. Do not alter clothing style."

If this is specifically what you're after, there's a fuller breakdown of free tools and settings in the piece on where to colourise old black and white photos with AI free, which covers a few alternatives alongside Gemini.

Couple, group and seasonal prompts

Group photos are where Gemini struggles most, because it has to keep multiple faces, poses and lighting conditions consistent at once. The fix is to be stricter, not looser.

• "Keep both people's faces, expressions and poses exactly as they are. Change only the background to a warm autumn garden with soft evening light, and match the lighting on both subjects to that new setting."
• "Keep the two subjects and their outfits unchanged. Add a festive string-light background and warm amber lighting, without changing their faces or body positions."

There are also occasion-specific sets worth copying wholesale rather than reinventing: the Karwa Chauth AI photo prompts for Gemini for couple edits, and a seasonal batch of Halloween photo edit prompts for Gemini if you're working around a specific date rather than a general portrait.

Step by step: how to run these prompts

1. Open Gemini (app or web) and start a new chat, or use Google's AI Studio if you want more control over the model version. In the Gemini app I couldn't find a Google page describing a separate image-edit model you have to switch on: you upload a photo, ask for the edit, and Gemini hands the job to its image model. If you want to pick the exact model version yourself, that's what AI Studio is for: choose the image model from the model list in the run settings panel before you upload, and the choice holds for that chat.
2. Upload your source photo first, before typing anything.
3. Paste a full four-part prompt (subject, change, style, what to keep) rather than a short instruction.
4. Review the result and make one correction at a time in the same chat thread, not a fresh prompt.
5. Download at full resolution. Gemini's free plan covers casual use, but if you're editing photos for client work or a product catalogue regularly, a paid Google plan raises the limit: Google's subscriptions page lists Google AI Plus at 2x the free plan's usage and Google AI Pro at 4x. Prices are set by country, so check that page in your own region for the current figure. What Google doesn't publish there is a fixed number of free photo edits a day: the paid plans are described as multiples of the free allowance, not as a count, and the free allowance itself isn't given as a number. In practice you find your limit when Gemini tells you you've reached it, so if you're editing a batch for a deadline, start early or check Google's plans page for current limits before you begin.
6. Keep your original file. Every edit is generative, not corrective, so if you want to try three different backgrounds, start each from the original, not from the previous edit.

Two things to know before you post anything. Google's developer docs say every image Gemini generates carries a SynthID watermark, an invisible marker that identifies it as AI-generated, so if you use an edited photo for a business, say it's been edited rather than passing it off as untouched. The same docs remind you to have the rights to any image you upload and not to generate content that deceives or harms, so stick to photos you took or have permission to use.

The part about faces that most guides skip over

Any generative edit can shift skin tone, flatten skin or hair texture, or change cultural and religious clothing details, even when you've asked it to keep them. Google doesn't publish figures on how often that happens or on which skin tones, so the only reliable check is your own: compare the result against the original at full zoom, every time, paying most attention to cheeks, hairline and any patterned fabric.

The workaround isn't glamorous. Be more explicit than feels necessary: name the exact skin tone ("deep brown, warm undertone, keep identical"), name the hair texture, and check the result against the original at full zoom before you use it anywhere public. If it's drifted, say so directly in the next message: "the skin tone has lightened, correct it back to match the original exactly." It usually takes one or two corrections rather than one. That check is the difference between a polished edit and one that quietly misrepresents the person in it.

The same honesty applies to consistency across a set. If you're building a batch of 40 similar portrait edits, for example the kind covered in Gemini AI photo prompts for girls with 40 edits, expect each one to need its own small correction pass. The model does not remember your style choices from one photo to the next unless they're in the same chat thread, so copying one prompt across 40 separate uploads will give you 40 slightly different interpretations of "studio lighting."

A note on naming the model

You'll see Gemini's image editing referred to as Nano Banana in a lot of prompt communities, and it's now Google's own term: its developer docs say "Nano Banana is the name for Gemini's native image generation capabilities" and list model versions under that name. Version names change as Google ships new ones, so check Google's image generation docs for the current list rather than trusting a version number quoted in a prompt guide, this one included. So a prompt written for "Nano Banana" is a prompt for Gemini's image editing, not a separate app. If you want a longer reference list beyond portraits and restorations, there's a broader set of Nano Banana prompts with 40 copy-paste Gemini image edits covering products, food, pets and travel photos too.

I keep every related walkthrough in the ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps. To write your own, try the free prompt generator.

Related reading: chatgpt photo editing prompts trending instagram.

Frequently asked questions

Do Gemini photo editing prompts work on any phone photo?

Yes, but quality in equals quality out. A blurry, badly lit original will still produce a blurry, oddly lit edit underneath the new background or outfit, because Gemini edits onto the existing detail rather than rebuilding the photo from scratch.

What should I type in Gemini to edit a photo?

Upload the photo, then use Google's own pattern: "Using the provided image, change only the [element] to [new description]. Keep everything else in the image exactly the same, preserving the original style, lighting, and composition." Then fix anything that drifts with one short follow-up per change.

Why does Gemini change my face slightly even when I ask it not to?

Every generation involves some reinterpretation, so small drift in skin tone, jawline or expression is normal, especially on darker skin tones where the model's training data is thinner. Always compare the result to the original at full zoom and correct it in the same chat if it's drifted.

Does Gemini watermark edited images?

Yes. Google's developer docs say all generated images include a SynthID watermark, which marks them as AI-generated even though you can't see it in the picture.

Is the free version of Gemini enough for regular photo editing?

For occasional personal use, yes. If you're editing photos daily for a business, a shop listing, or client work, a paid Google AI plan raises the limit: Google AI Plus is $4.99 a month for 2x the Free usage limits and Google AI Pro is $19.99 a month for 4x (Google's subscriptions page, checked in November 2026).

What is Nano Banana?

Google's docs describe Nano Banana as the name for Gemini's native image generation and editing, which works conversationally with text and images. It isn't a separate app; it's what runs when you ask Gemini to edit a photo.

Can I use these prompts for commercial product photos?

Yes, once it passes three checks: the colour matches the real item side by side, any logo or label text is unchanged, and no parts have been invented (an extra strap, a second button, a handle that isn't there). Keep your original product shot untouched, be exact about background and lighting changes, and always check colour accuracy against the real product before publishing, since generative edits can subtly shift shade and texture.

Primary sources

• Anthropic's own documentation
• the NIST AI Risk Management Framework