18 Nano Banana Face Swap Prompts for Hyper-Realistic Results

Face swapping is one of the hardest things to ask an image model for, because the failure mode is so specific: the result can be technically flawless and still not look like the person. Nano Banana face swap prompts solve that by naming what must stay fixed, rather than hoping the model works it out.

Below are 18 detailed prompts across professional headshots, cinematic scenes, artistic styles, travel shots, and — the section most guides skip — fixing a swap that came back wrong. Run them wherever you use Nano Banana, or in Chat Smith, an AI assistant that keeps the image models and the chat models in one place.

Nano Banana is the codename for Google's image generation engine, available in a standard version and a higher-resolution Nano Banana Pro variant. What makes it suited to this task is character consistency: the ability to carry the same identity across different scenes, lighting conditions and art styles. You can run the standard Nano Banana model for drafts and switch to Pro when you need print resolution.

Older approaches essentially glued a face onto a head, which is why the results always looked pasted on. Newer models adapt the lighting, skin texture and apparent bone structure to the target scene instead — but only when the prompt tells them what the target scene's lighting actually is. That is the difference the prompts below are built around.

Every prompt below follows the same five-part shape. Learn it and you can write your own for any scene — the same logic applies to any image to image workflow.

Label your references. "The face in Image 1, the scene in Image 2." Multi-image reference works well, but only when you say which image is which.

Lock the identity by naming features. Jawline, eye shape and spacing, nose bridge, eyebrow shape. "Make it look like them" is not an instruction; a list of features is.

Match the light before the style. State the direction, hardness and colour temperature of the light in the target scene. Most failed swaps fail here rather than on the face itself.

Ask for skin, not smoothness. Pores, grain, slight asymmetry, subsurface scatter. The plastic look comes from a model given no texture instruction at all.

Specify the output. Resolution, aspect ratio, depth of field. If you need print quality, say so in the prompt rather than upscaling afterwards.

One thing worth stating plainly before the prompts: only swap a face you have the right to use. Your own is fine. Someone else's — a colleague, a public figure, a photo taken from social media — needs their consent, and in many places using it commercially or in a way that misleads is unlawful regardless of how convincing the result is. Most platforms prohibit it in their terms as well. Everything below assumes the face in Image 1 is yours or one you have permission to use.

Four prompts for profile photos and team pages, where the likeness matters more than the styling. If you want a guided flow instead of writing the prompt yourself, an AI headshot generator or an AI portrait generator covers the same ground.

1. The executive close-up

Using the face from Image 1, swap it onto the subject in Image 2. Keep the jawline, eye spacing and nose bridge identical to Image 1. Expression: a warm, closed-mouth smile. Lighting: soft-box key at 45 degrees camera left, subtle fill, shallow depth of field around f/1.8. Render realistic skin grain and visible pores — do not smooth the skin.

2. The keynote speaker

Swap the face from Image 1 onto the speaker on stage in Image 2. Hold the jawline, eye shape and eyebrow arch exactly as they appear in Image 1. Lighting: cool blue stage wash from the front, warm rim light on the shoulders and one side of the jaw. Mid-shot, slight upward camera angle, shallow background blur.

3. The minimalist portrait

Transfer the facial identity from Image 1 onto a subject in a beige turtleneck against a charcoal seamless background. Preserve bone structure and the asymmetry of the original face. Single soft light from above and slightly right. Emphasise skin texture and catchlight in the eyes. Square crop, high resolution suitable for print.

4. The golden hour office shot

Replace the subject's face in Image 2 with the face from Image 1, keeping eye colour and lip shape unchanged. The subject is seated in a glass-walled office at sunset, warm low-angle light hitting the left side of the face and leaving the right in soft shadow. Include realistic lens flare and window reflection, but keep the face fully legible.

Four prompts where the scene does the heavy lifting. The risk in stylised environments is identity drift, so each of these names the features to hold constant while everything around them changes.

5. The cyberpunk rebel

Insert the face from Image 1 into a rain-soaked neon street at night. The subject wears a high-collar leather jacket. Critical: the face must be lit by the environment — magenta from the left, cyan from behind, wet skin catching both. Keep jawline and eye spacing identical to Image 1. Shallow depth of field, background signage out of focus.

6. The high-fantasy knight

Swap the identity from Image 1 onto a warrior in ornate silver armour standing in a misty forest at dawn. Preserve exact eye colour, eyebrow shape and the original nose profile. Cool diffused light through fog, faint warm sunrise behind. Add cold-flushed skin and light condensation. Cinematic colour grade, no beautification of the face.

7. The 1940s noir detective

Apply the face from Image 1 to a detective in a rainy alleyway, hat brim casting a hard shadow across the upper face. Style: film noir, black and white, high contrast chiaroscuro with a single hard key light. The eyes must stay visible and recognisable despite the shadow. Heavy vintage film grain and slight halation on the highlights.

8. The space explorer

Swap the face from Image 1 into a detailed astronaut helmet. Preserve facial bone structure precisely. The visor should carry a faint reflection of a distant galaxy without obscuring the face, and the interior helmet lighting should come from below, tinted cool white. Add breath condensation on the inside of the visor near the mouth.

Four prompts for stylised output, where the challenge reverses: the style must transform everything except recognisability. Useful for profile images, and a close cousin of what an AI avatar generator does in one click.

9. The Renaissance oil painting

Reinterpret the face from Image 1 as the subject of a 16th-century oil portrait. Keep the likeness — bone structure, eye spacing, mouth shape — but adapt skin tones to a sfumato technique with soft transitions. Dark background, single window light from the left. Heavy impasto texture on the clothing, visible craquelure across the varnish.

10. The 90s grunge polaroid

Swap the face from Image 1 onto a subject in a grainy lo-fi 90s Polaroid. Direct on-camera flash, slightly overexposed forehead and cheeks, hard shadow behind the head. Mottled blue-grey mall studio backdrop, subtle motion blur, colour shifted warm and slightly faded. Preserve recognisable features despite the exposure.

11. The flat vector illustration

Convert the facial identity in Image 1 into clean minimalist vector art: bold consistent line weight, flat colour fills, no gradients. Reduce the face to its most identifying features — the eyebrow shape, the jaw angle, the hairline — so it stays recognisable at avatar size. Limited palette of four colours, transparent background.

12. The marble bust

Using Image 1 as reference, render a hyper-realistic white marble bust. The features must be carved with precision so the likeness survives the loss of colour and texture. Include subtle veining in the stone, a hairline crack near the base, and slight chipping at one edge. Museum lighting from above, neutral grey background.

Three prompts where natural light is doing the work. Outdoor scenes are unforgiving — if the light on the face does not match the light in the scene, the eye catches it instantly.

13. The Amalfi Coast morning

Place the face from Image 1 onto a person having an espresso on a balcony above a Mediterranean coastal town. Bright late-morning sun from above and behind, bounce light from the white balcony wall filling the shadows under the chin. Saturated blues and terracotta. Slight squint from the brightness — keep it natural rather than posed.

14. The Tokyo night walk

Swap the face from Image 1 onto a subject walking through a busy neon-lit crossing at night. Fast shutter freezing the subject, mild motion blur in the crowd behind. Mixed colour lighting from signage above and screen glow from the left. Keep the eye shape and jawline locked to Image 1 despite the colour cast.

15. The mountain summit

Swap the face from Image 1 onto a hiker on a snowy peak. Harsh high-altitude sun with strong bounce from the snow below, filling the underside of the chin and nose. Add cold-flushed cheeks and nose, chapped lips, wind-blown hair, and faint goggle tan lines. Preserve the original bone structure exactly.

Most face swaps fail in one of three ways, and each has a specific fix. Diagnose before you regenerate — rerunning the same prompt with different wording rarely helps. If the composition is right and only the resolution disappoints, an AI image enhancer is the quicker route.

16. The identity drift fix

The swap in this result no longer looks like the person in Image 1 [attach both]. First, tell me which specific features drifted — jaw width, eye spacing, nose bridge, lip shape, hairline. Then regenerate with those features locked to Image 1, while keeping the scene, pose and lighting of the current result unchanged.

17. The pasted-on lighting fix

This face swap looks pasted on [attach the result]. The problem is lighting, not the face. Analyse the direction, colour temperature and hardness of the light in the scene, then regenerate so the shadows on the nose, under the chin and in the eye sockets match it. Add any rim light the background source would realistically cast, and match the skin's colour cast to the environment.

18. The plastic skin fix

This result looks over-smoothed and uncanny [attach it]. Regenerate with realistic skin: visible pores, natural asymmetry between the left and right sides of the face, existing blemishes and lines kept rather than removed, fine facial hair, and subsurface scattering at the ears and nostrils. Match the film grain of the surrounding image. Do not beautify or slim the face.

Face swapping is iterative by nature: the first result is rarely the one you keep, and the useful version of this workflow is generate, diagnose, adjust one variable, regenerate. Chat Smith lets you save your working prompt as a template, keep the identity-lock clause consistent across every variation, and launch it again without retyping the whole thing.

It also gives you the full model library in one place, so you can draft on the standard model and switch to Pro for the final render, or compare how two engines handle the same reference face. The rest of the AI image tools sit alongside it for background removal, upscaling and the finishing work.

Start with prompt 1 on a photo of yourself, then change exactly one variable — the lighting direction — and run it again. Seeing what that single change does to the result teaches you more about these models than any prompt list, including this one.