Nano Banana 2.1 Prompts: Same Face, New Scenes, Clean Edits
Ready-to-use Nano Banana 2.1 prompts that keep one character's face across new scenes, outfits and product shots, with the reference setup for every recipe.
The Pattern Behind Every Recipe
Nano Banana 2.1 works best when images carry the information and the prompt gives instructions. Every recipe here has three parts:
Name what to keep, by number. References count in the order you attach them, so “the woman from image ref 1” points at one picture, while “a woman” invites a new face.
Say what changes, and only that. One change per prompt keeps the rest stable.
End with the look and format. Light, camera feel and aspect ratio.
You can attach up to 14 references, with room for four consistent characters and ten objects. Every recipe runs in the Image Lab for 12 credits per 2K image, references included.
Start With an Anchor Portrait
Make one clean portrait first and use it as the face reference for everything after.
Prompt: Head-and-shoulders portrait of a fictional 32-year-old woman with shoulder-length auburn hair and light freckles, neutral expression, looking straight into the camera, plain light-gray wall, soft even daylight, sharp focus on the eyes, natural skin texture Format: 4:5 at 2K
Front-facing, even light and a plain background give the model a face it can read. Attach that same file every time; swapping in a newer output lets small changes compound. Use fictional characters, or real people only with their consent.
Same Face, New Places
Face into a real photo. Image ref 1 = a photo you took or have the rights to, with the pose and setting you want; image ref 2 = your anchor. Prompt: Keep image ref 1 exactly as it is, including the pose, clothing, room and light, and give the person the face and hair of the woman in image ref 2
A real photo keeps real light and framing. The playbook's first-frame lesson covers this step and the enhance pass after it.
New scene from scratch. Image ref 1 = the anchor; image ref 2 = a location photo (optional). Prompt: The woman from image ref 1 reading a paperback by a train window in late-afternoon sun, navy raincoat, warm side light on her face, candid phone photo, 9:16
Two characters together. Image ref 1 = the first anchor; image ref 2 = the second. Prompt: The man from image ref 1 and the woman from image ref 2 sharing a small table at a street café, both laughing, the man sitting closer to the window, overcast daylight
Tie positions to objects (“closer to the window”), not just left and right, which Google lists as a known mix-up.
Recipes for Edits
Outfit change. Image ref 1 = your character; image ref 2 = the garment, ideally laid flat. Prompt: Dress the woman in image ref 1 in the knit cardigan from image ref 2; same face, hair, pose and background, only the clothing changes
Product placement. Image ref 1 = the product; image ref 2 = the scene. Prompt: Place the glass perfume bottle from image ref 1 on the stone ledge in image ref 2, morning light from the window, crisp unchanged label, soft shadow beneath it, 4:5
Check the label against the real product, and for packaging with small print, test GPT Image 2.5 too.
Enhance pass. Image ref 1 = a frame that looks soft or over-smoothed. Prompt: Sharpen this photo without changing who is in it: same face, hair, expression, clothing and framing; restore fine skin texture and natural light
Text on a cover. Image ref 1 = your character. Prompt: Vertical cover image, the woman from image ref 1 on the right holding a folded paper map, surprised, bold white text ‘LOST IN LISBON’ on a teal block on the left, high contrast, 9:16
Keep text to one short, large line in quotes; Google's model card says small text and long paragraphs are still weak. For dense layouts, see when to switch to Pro.
Chain Edits Without Losing the Face
Most face drift comes from feeding each output into the next edit. A safer routine:
Change one thing per generation, and say what stays: “same face as image ref 1”.
Keep the anchor attached. When you choose Edit Image on a result to send it back to the Image Lab, add the original anchor too, and point the face at the anchor, not the last output.
Check before moving on. Compare eyes, jawline and hairline with the anchor, and redo any step that slipped.
Watch the pose. Now and then the model keeps the input pose. Name the new one plainly (“standing, arms crossed”) or attach a pose reference.
From Still to Series
Press Animate on a finished image to open it in the Video Lab with Seedance 2.5; a 5-second clip costs 50 credits at 480p. Consistent AI characters goes deeper into the one-face-reference method, the free AI OFM course builds a fictional adult persona from its first image up to motion-control, lip-sync and sketch clips, and Nano Banana 2.1 explained covers what's new.
Frequently Asked Questions
What's the best way to keep a consistent character in Nano Banana 2.1?
+
Make one front-facing anchor portrait and attach that exact file to every prompt. Refer to it by number, list what must not change, and change one thing per generation, checking each result against the anchor.
What's the reference image limit in Nano Banana 2.1?
+
Fourteen images per request. Google's documentation says that covers up to four consistent characters and ten objects. In Spicy Studio's Image Lab, references add no credits: a 2K image is 12 credits either way.
Can Nano Banana 2.1 do a face swap on a photo?
+
Yes. Attach the photo as image ref 1 and the face as image ref 2, and ask it to keep image ref 1 but use the face and hair from image ref 2. Only use photos you have rights to, with fictional faces or people who have agreed.
Why does my character's face change after a few edits?
+
Each edit probably started from the previous output, so small changes added up. Re-attach the original anchor at every step and check each result; Google's model card notes that reference-to-output consistency isn't always perfect.
Want to run this yourself? Every new account starts with free credits.
Try these prompts in the Image Lab