Gemini AI photo prompt: copy-paste prompts for editing your own photo

Start here: a prompt that edits your photo and keeps your face

Open the Gemini app, choose Create images from the tools menu, upload your photo and paste this. It changes one thing and tells Gemini to leave the rest alone, the local-edit pattern from Google's developer guide.

What to change: the bracket. Send one change per message. Where it breaks: if the face still drifts, describe it (hair color, eye color, glasses, beard). Google says to describe critical details like a face alongside your edit request.

The six-part formula behind every Gemini prompt

Google's Gemini page suggests a simple pattern: create an image of a subject, an action and a scene, then add detail. For a photo you own, six parts cover it.

1. Subject and what must stay the same. Who is in the photo and which features are locked: face, hair, skin tone, expression.
2. Scene. Where the person is and what is happening.
3. Lighting. Soft window light, golden hour, a studio softbox.
4. Camera or lens. Google says photographic terms such as wide-angle shot or low-angle perspective control the composition. For portraits, add a lens such as 85mm.
5. Style. Photorealistic, 35mm film, watercolor. A style transfer re-renders the subject, says Google's tips post, so pick a painterly style only when you want that.
6. Format. Write the shape in, as Google's example does with a 2:3 aspect ratio: 4:5 for portraits, 16:9 for widescreen, 1:1 for squares.

Write full sentences, not a string of tags. Fill in this template, or paste rough notes into Gemini and ask it to turn them into a prompt with these six parts. Keep to one scene per request.

Prompts for the portrait looks people ask for most

Change: outfit, backdrop, format. Breaks: smoothed skin. See the face fixes below.

Change: the scene and colors. Breaks: heavy style words can re-render the face. Drop the color grade first and add it in a second message.

Change: the decade. Google's own retro example is one line: "Turn this into a retro-style mall studio portrait." Breaks: strong grain softens the eyes.

Change: people, location, light. Breaks: Google's developer documentation limits how many people's likeness the model keeps in one workflow, so keep the group small and check every face.

Change: season, setting, clothes. Breaks: a new season changes the light on the face too, so name the light.

Product photos and old-photo restoration

The product prompt adapts Google's studio product template; Google lists photo restoration among the edits Gemini handles in conversation.

Change: surface, angle, key detail. Breaks: labels and logos. Google says Gemini may misspell words, so read every letter.

Change: delete the last sentence to stay black and white. Breaks: tiny or blurry faces. Restoration fills in missing detail, so compare each face with the original.

Why the face changes, and how to bring it back

Google's tips post calls character consistency a strength but not perfect. Most changed faces have a fixable cause.

• The prompt never said what to keep. Add the lock line: keep my face, features, skin tone and expression exactly the same.
• Too much at once. Google's developer guide says that when an image is mostly right, edit it with the next small request instead of generating a new one.
• An art style. A style transfer re-renders the subject. For a real person, stay with photorealistic.
• The model setting. Google says the Flash-Lite setting (Nano Banana 2 Lite) is not designed for multiple edits, so switch the Gemini model first. On a paid Google AI plan, Redo with Pro uses Nano Banana Pro, which Google describes as more consistent.
• Plastic skin. Drop words like flawless and airbrushed. Ask for natural skin texture; Google's guide favors describing what you want over what you don't.

Extra fingers, garbled text and the wrong shape

• Extra or bent fingers. Google's pages do not single out hands, so there is no documented switch. The documented tool is the local edit: Using the provided image, change only the left hand so it has five fingers in a relaxed pose. Keep everything else exactly the same. If two tries fail, use Redo with Pro.
• Wrong or misspelled text. Google says Gemini may misspell words or struggle with complex typography. Put the exact words in quotation marks, keep them short and name the font style. Google's developer documentation adds that text works best when you generate the words first, then ask for the image.
• Wrong aspect ratio. Google's tips post says the model struggles to hold aspect ratios. Write the ratio into the prompt and crop your upload to that shape first, since Google's developer documentation says output size follows the input image by default. Still off? Ask for a resize.

The tips post dates from August 2025, so treat its limits as things to check.