A Gemini AI face consistency prompt attaches your approved images, gives each one a role, asks for one change and ends with a keep line naming the face, hair and proportions. This pack has 30 fill-in templates for Nano Banana, grouped by task with a drift-risk note for each group. They follow Google's documented prompt patterns and are starting points to adapt, not scored recipes.
A Gemini AI face consistency prompt has three working parts: your approved reference images with a role for each, the one change you want, and a keep line that names the face, hair and proportions. The 30 templates below apply that structure to the jobs a persona account repeats every week: building the reference set, changing a pose, swapping an outfit, moving to a new scene, making an expression sheet and putting two characters in one frame.
Templates follow Google's documented patterns, checked October 2026: the Gemini API image generation docs and Nano Banana 2.1 model page, the Gemini Apps image help page, Google's Nano Banana Pro prompt tips and Google Cloud's Nano Banana prompting guide. We did not score them in a test; output varies by model and reference set.
Prompts come second. The reference set comes first, which is why templates 1 to 5 build one. For where Gemini sits among the other methods, see our consistent character AI guide; this page is the copy-and-paste layer for people who have already chosen Nano Banana.
The Gemini AI face consistency prompt, part by part
Every template in the pack is a cut-down version of this five-part prompt. You rarely need all five parts; you always need the references and the keep line.
[REFERENCES] Using the attached images of [Name] [ROLE] (image 1 is her face, image 2 is the pose to copy), [CHANGE] create a new photo: [the one thing that is new]. [CAMERA] [Shot type], [light], [lens or angle]. [Aspect ratio]. [KEEP] Keep her face, hair and proportions exactly as in image 1.
| Part | What it does | Google's wording or limit |
|---|---|---|
| References | Puts the face in front of the model instead of describing it | Nano Banana models mix up to 14 reference images |
| Role | Stops the model borrowing a face or outfit from the wrong image | "When using uploaded images, clearly define the role of each" |
| Change | Limits the edit to one thing you can judge | "Change only the [specific element] to [new element/description]" |
| Camera | Fixes framing so the face is large enough to hold | "Control the camera" with photographic language |
| Keep | Names what must not move | "Ensure that the features ... remain completely unchanged" |
- 1Attach the references first
Your approved hero plus the view closest to the angle you want. Fewer, cleaner images beat a pile of similar ones.
- 2Replace the brackets
Swap [Name], [outfit], [place] and [light] for your own. Leave the keep line as written.
- 3Ask for one change
Pose or outfit or scene. If you need all three, do them in three turns.
- 4State the aspect ratio
By default the model matches your input image or returns a square, so name the ratio your feed uses.
- 5Compare with the hero
Check eyes, nose, jaw and hairline at full size before you keep the image.
- 6Feed keepers back in
Google says to include previously generated images in later prompts. Add each approved view to the set.
Which Nano Banana model to run them on
In the Gemini app, the model setting decides which image model answers. Google's help page says Flash or Pro uses Nano Banana 2, which accepts multiple reference images, and Flash-Lite uses Nano Banana 2 Lite, which is not designed for multiple image references or multiple edits. Paid subscribers can redo an image with Nano Banana Pro. In the API, Nano Banana 2.1 is the primary model.
- Character images: up to four on Nano Banana 2.1 and Nano Banana 2, up to five on Nano Banana Pro, inside a total of 14 reference images.
- Characters per workflow: Google's limitations list gives character resemblance for up to four characters.
- Avoid Lite for this job. Templates 6 to 30 all assume reference images and follow-up edits.
Nano Banana prompts to keep the same face: 30 templates
Six tasks, five templates each. The drift-risk label is our reasoning from how much each prompt asks the model to invent, not a measured rate. The examples use "she"; change pronouns and wardrobe to suit an adult persona of your own.
| Task | Templates | Drift risk |
|---|---|---|
| Build the reference set | 1 to 5 | Lowest |
| Pose change | 6 to 10 | Low to medium |
| Outfit swap | 11 to 15 | Low for the face |
| Scene and lighting | 16 to 20 | Medium |
| Expression sheet | 21 to 25 | Medium to high |
| Group scenes | 26 to 30 | Highest |
1 to 5: Build the reference set
Drift risk: lowest. Plain background, even light and one angle per image leave the model little to invent. Do these first: every later template leans on them.
1. Create a photorealistic studio portrait of [character line]. Front-facing, neutral expression, soft even light, plain light-grey background, head and shoulders. Aspect ratio 4:5. 2. A studio portrait of this person against the same light-grey background, three-quarter view looking left. Keep the face, hair and skin exactly as in the attached image. 3. A studio portrait of this person against the same background, in profile looking right. Keep the facial features, hairline and hair length exactly as in the attached images. 4. A full-body studio photo of this person standing, arms relaxed, feet visible, plain background. Same face, hair, build and proportions as the attached images. 5. A studio portrait of this person seen from slightly above, looking up at the camera. Keep the facial features exactly as in the attached images.
6 to 10: Pose change
Drift risk: low to medium. Risk rises as the head turns away from the angles your references show. Attach the view nearest the new pose.
6. Using the attached images of [Name], create a new photo: she sits on stone steps, elbows on knees, looking at the camera. Keep her face, hair and outfit exactly the same. 7. Using the attached images of [Name], create a new photo: she walks toward the camera on a city pavement, mid-stride, full body visible. Same face, hair and build as the references. 8. Image 1 is [Name]. Image 2 is a pose reference only. Put [Name] in the pose from image 2. Take nothing else from image 2: not the face, hair, clothes or background. 9. Using the provided image, turn her head to look out of the window on the left, three-quarter view. Keep everything else in the image exactly the same. 10. Using the attached images of [Name], create a close-up: she rests her chin on her hand. Keep the facial features, the mole under her left eye and the hair unchanged.
11 to 15: Outfit swap
Drift risk: low for the face. The face is not being asked to change, so it usually has the least reason to. Watch body proportions and the hairline, which a new collar or hairstyle can pull.
11. Using the provided image, change only the outfit to [outfit]. Keep everything else in the image exactly the same, preserving the face, hair, pose, lighting and composition. 12. Image 1 is [Name]. Image 2 is a jacket. Let the person from image 1 wear the jacket from image 2. Ensure her face and features remain completely unchanged. 13. Create a full-body photo of the person in the attached images wearing [outfit], standing in [place]. Same face, hair, build and height impression as the references. 14. Using the provided image, add thin gold-rimmed glasses. Make them sit naturally on her face. Do not change the eyes, brows or anything else. 15. Using the provided image, change only the hairstyle to a low ponytail. Same hair colour and hairline. Keep the face exactly the same.
16 to 20: Scene and lighting
Drift risk: medium. New light redraws the shadows that define a face, and wide shots shrink it. Check wide shots zoomed in.
16. Using the attached images of [Name], create a photorealistic medium shot of her in [location]. [Light]. Shot at eye level with a 50mm lens. Same face, hair and proportions as the references. 17. Using the provided image, change only the background to [place]. Keep the person exactly the same, and adjust the light on her to match the new scene. 18. Keep everything the same, but change the time to dusk with warm street lighting. 19. Using the attached images of [Name], create a photo of her at a restaurant table lit by one candle. Keep her facial features recognisable; do not smooth or reshape the face. 20. A wide shot of [Name] from the attached images standing on a beach at sunrise, full body, small in the frame. Same build, hair and outfit as the references.
21 to 25: Expression sheet
Drift risk: medium to high. A big expression moves every landmark you use to recognise the face. Make each expression as its own image from the neutral portrait.
21. Keep everything the same, but change her expression to a soft closed-mouth smile. 22. Keep everything the same, but change her expression to a laugh with teeth visible. Keep the eye shape, nose and jawline the same. 23. Keep everything the same, but change her expression to serious: lips together, direct eye contact. 24. Keep everything the same, but change her expression to surprised: eyebrows raised, mouth slightly open. 25. Create a 2 by 2 grid of studio portraits of this person on a plain background: neutral, smiling, laughing, serious. Same face, hair and framing in every panel.
26 to 30: Group scenes
Drift risk: highest. Each extra person is another identity to hold and another face to mix up. Give every character their own image and name them by image number.
26. Image 1 is [Name A]. Image 2 is [Name B]. Create a photo of both at a cafe table, [Name A] on the left, [Name B] on the right. Keep each person's face and hair exactly as in their own image. 27. Using the attached images of [Name], create a photo of her in a busy market. Other people are in the background and out of focus. Only [Name] is in sharp focus. 28. Image 1 is [Name]. Image 2 is the product. She holds the product at chest height, label facing the camera. Keep her face unchanged and the product's shape, colour and label exact. 29. A group photo of these people standing in a row in an office, relaxed smiles. Keep every person's face consistent with their own reference image. 30. Using the provided image of [Name], add [Name B] from image 2 standing behind her right shoulder. Do not change [Name]'s face, pose or the lighting.
Template 1 needs a character line: a fixed description of a fictional adult. Our AI influencer prompt pack explains how to write one and has 100 scene lines to drop into the bracketed slots, and the free AI influencer prompt generator builds the line for you.
Gemini prompt for a consistent character: reading the drift risk
Two things decide how likely a face is to slip: whether a reference shows the angle you are asking for, and how many things the prompt changes at once.
How to adapt the templates
- Work in turns. Google calls multi-turn conversation the recommended way to iterate on images. Templates that begin "Keep everything the same, but" are follow-ups to the image in the chat.
- Return to the originals. After a few edits in one thread, start a new chat with your approved references. Editing a drifted image carries the drift forward.
- Describe what you must keep. Google's detail-preservation template says to describe critical details, such as a face, in great detail alongside the edit request. Name the mole, the freckles, the hair parting.
- Say it positively. There is no negative prompt field. Google suggests semantic negative prompts: describe the scene you want instead of listing what to leave out.
- Add a pose reference for complex poses. That is Google's own advice in its character consistency example, and template 8 does it.
- Same girl but at the beach in a red dress, smiling
- Make her look better
- Put these two people together
- No weird face, no plastic skin
- Do it again but different
- Change only the background to a beach at golden hour. Keep the person exactly the same
- Keep everything the same, but soften the light on her face
- Image 1 is Mara, image 2 is Jon. Mara on the left, Jon on the right
- Natural skin texture with visible pores. Keep her facial features unchanged
- Keep everything the same, but turn her head to a three-quarter view
If Gemini is one of several tools you use, the same structure carries over: our ChatGPT consistent character guide uses a description block built on the same change-then-keep idea, and consistent characters without a LoRA ranks the no-training methods side by side. Prompts take a persona a long way; the AI Influencers program goes deeper with lessons on prompt engineering, creating the persona, LoRA training for consistent faces and photorealism.
Before you publish an image
- Eyes, nose, jaw and hairline match the hero image at full size
- Fixed marks (mole, freckles, scar) are present and on the correct side
- Hair colour, length and parting match the character line
- Hands, teeth and ears look right at 100 percent zoom
- The face is the persona's own and does not resemble a real person
- The prompt and the references used are saved with the image
- The post carries the AI label the platform requires
Two rules sit above the checklist. Use an original persona or your own face: Google's docs remind you to have the necessary rights to any image you upload, and its Prohibited Use Policy bars impersonating an individual without explicit disclosure in order to deceive. And expect the watermark: Google states that all generated images include SynthID. For platform labelling, see Instagram's AI label rules, and when a face keeps slipping whatever you type, the AI face consistency guide covers the causes beyond the prompt.
Gemini face consistency prompts: FAQ
What is the best Gemini prompt to keep the same face?
One that attaches your approved images, gives each a role, asks for a single change and ends with a keep line. For example: "Using the provided image, change only the outfit to a navy blazer. Keep everything else exactly the same, preserving the face, hair, pose and lighting." Google's own editing templates use the same change-only-then-keep structure.
How many reference images should I give Gemini for one character?
Three or four views of the same face is a practical set: front, three-quarter, profile and full body. Google's docs say Nano Banana 2.1 and Nano Banana 2 take up to four character images to maintain character consistency, and Nano Banana Pro up to five, within a total of 14 reference images. Checked October 2026.
Why does Gemini change the face when I edit an image?
Usually because the prompt changed several things at once, asked for an angle no reference shows, or left out what to keep. Google's prompting tips also state that character consistency across edits may vary. Ask for one change per turn, restate what must stay the same, and when a face slips go back to your original references instead of editing the drifted result.
Do these prompts work in Nano Banana Pro and through the API?
Yes. They are plain-language instructions, so they work in the Gemini app, Google AI Studio and the Gemini API. Check the model: Google describes Nano Banana 2 Lite as not optimized for multiple reference inputs or multi-turn editing, so use Nano Banana 2.1, Nano Banana 2 or Nano Banana Pro for persona work.
Does Gemini have a negative prompt for faces?
There is no negative prompt field. Google recommends what it calls semantic negative prompts: describe the scene you want in positive terms instead of listing what to exclude. For a face, write "natural skin texture with visible pores" rather than "no plastic skin", and state what to preserve, such as "keep her facial features exactly the same".
Can I use a real person's photo with these prompts?
Use your own face or an original AI persona. Google's docs remind you to have the necessary rights to any image you upload, and its Prohibited Use Policy bars impersonating an individual without explicit disclosure in order to deceive. Editing images in the Gemini app is limited to users aged 18 or over, and every generated image carries a SynthID watermark.
Thirty prompts hold a face. A persona needs the rest of the system.
AI Influencers, included in All Access, covers what comes after the prompt: creating the persona, LoRA training for consistent faces, photorealism and video, with the other three programs, weekly coaching and the private community in one subscription.
Get the character line written for you
Build the fixed character line template 1 needs with the free prompt generator, and follow the free Telegram channel for new Nano Banana releases and prompt patterns.