Profile pictures
Make a version of your selfie for Instagram, TikTok or Discord.
Free · No sign up for your first image
Nano Banana 2.1: What's New, Prompts & How to Try It
Change what you want. Keep what matters.
What changed in Google's Nano Banana 2.1, prompts that show it off, and an editor to try the same edits. This page currently runs Nano Banana 2 — we'll switch to 2.1 here as soon as it's available to us.
No sign up
No added visible watermark
Auto-delete in 24h
Never used for training
1Upload image
2Choose change
3Generate
Step 1 · Add your image
Drop your image here
Drag & drop, paste, or choose from your device
Upload image
JPGPNGWebP≤ 10 MB
Private · auto-deleted in 24h
No photo handy? Try a sample
What do you want to change?
Pick one direction. Hold Frame will protect the important parts automatically.
We’ll detect what matters after you upload, then protect it automatically.
2 Choose one you can refine it below
Advanced settings Optional
Detected mode after upload
Keep protected recommended for this image
Change strength
Balanced — clear change, identity and layout stay stable.
Aspect ratio
✦ 1 free result as a guest · failed generations never count
Real before & after
Example edits with these prompts (made with Nano Banana 2)
Inputs are AI-generated or our own product photos. Outputs came from this page's editor, which currently runs Nano Banana 2, not 2.1.
The control other tools hide
You don't pick a model. You pick what survives.
Most img2img tools give you one abstract “strength” slider. We turn it into decisions a human actually makes.
Keep — hard constraints
Locked items become explicit model instructions — and a region mask when you paint one. The AI regenerates the whole image using your photo as the reference, so compare locked details with the original before you publish.
- Face — it's still the same person. Identity reference plus face-region protection.
- Subject — the product, pet or object keeps its shape, logo and key details.
- Composition — pose and layout stay put via structure and depth control.
- Text & Logo — the model is told to keep lettering and logos as they are; fine print can still change, so check it.
Change — creative freedom
One tap per intent. Each chip compiles into a structured instruction with the right negative constraints.
- Style — anime, watercolor, film, 3D, oil paint and 40+ more.
- Background — a new scene behind an untouched subject.
- Outfit — swap clothing while the person stays the person.
- Lighting — golden hour, studio, neon night, overcast.
Subtle
Light touch: tone, light, small edits. Routed to precise editing models.
Balanced
The default. Obvious change with stable identity and layout.
Creative
Big reimagining. Structure loosens — fine details may drift, on purpose.
Three steps
From photo to finished image in under a minute
Upload & auto-detect
Drop any JPG, PNG or WebP. We recognize portraits, products, sketches and interiors, and pre-lock what usually matters for that kind of image.
Lock, describe, set strength
Toggle what to keep, tap what to change, pick Subtle, Balanced or Creative. No denoising values, no seed numbers, no model menus.
Compare, download, iterate
Slide before/after, download free with no added visible watermark, then keep editing the result with plain language — no re-upload needed.
One-tap starts
Prompt templates that fill the tool for you
Tap any template — it configures keep, change and strength, then scrolls you back to the studio.
Complete guide
How to try Nano Banana 2.1-style edits online
Nano Banana 2.1 is Google's newest Flash-tier image model. Below: what changed, prompts that work, and an editor to try the same edits — currently running Nano Banana 2.
In short
Nano Banana 2.1 is Google's image generation and editing model (official model code gemini-nano-banana-2.1), listed as stable in Google's Gemini API docs on October 6, 2026. Google describes it as an update to Nano Banana 2 with better image quality and realism at 1K, 2K and 4K, sharper text and infographic layouts, up to 14 reference images and adjustable thinking levels. You can use it in the Gemini app and in AI Mode in Google Search. On this page you can try the same kinds of edits online — they currently run on Nano Banana 2 (Gemini 3.1 Flash Image), not 2.1; the result label says which model made it. People also search for it as nano banana 2.1 AI, nanobanana 2.1, nanobanana21, nano banana2.1, nano banana 2 1, nano banana 2.1 free, nano banana 2.1 online, nano banana 2.1 prompts.
Facts checked October 10, 2026.
Nano Banana 2.1: what you get
| Free to try | First standard (about 1K) result free as a guest — no sign-up, no credit card, no app download. Failed generations don't count. |
|---|---|
| After the free result | 5 credits per standard image. One-time packs: $6 for 50 credits (10 images) or $12 for 125. Plans: $9/month for 100, $19/month for 250; credits carry over. |
| Watermark | No visible watermark is added. The model provider's invisible provenance mark (SynthID) is not removed. |
| Original pixels kept? | For background changes, yes: with "Keep subject pixels exactly" (on by default for product photos, one tick for people) the original product or person is cut out in your browser and pasted back pixel-for-pixel; only the background is AI — same price. Other edits are regenerated by the image model (Google Gemini 3.1 Flash Image) with your photo as the reference, so compare locked details. |
| Best for | Quick, controlled edits from one photo when you want to try several directions before committing. |
| Not the right tool for | Pixel-exact compositing (use layer masks in Photoshop) or editing dozens of images in one batch. |
What you can make
The trend
Post the before/after — the side-by-side is what makes the trend work.
Friends and parties
With permission, make versions of your friends for a group chat or invite.
How to get a controlled result
A three-step workflow for Nano Banana 2.1
Use one clear face
A sharp, well-lit selfie or waist-up photo works best. One person per photo.
Pick a style
Tap a style or write your own. Balanced keeps you recognizable; Creative goes further.
Review and retry
If the face drifts, lower the strength and keep Face locked.
Prompt examples
Copy one as a starting point, then change only the details that matter to your image.
- turn this photo into a 1970s film still: warm film grain, soft halation, period-correct styling; keep the person's face and pose
Common problems and fixes
Use these checks before increasing change strength or adding more instructions.
It doesn't look like me. Switch to Balanced or Subtle strength and keep Face locked.
The engine declined the photo. Some photos are declined by the safety checks; declined or failed generations don't use your free try or credits.
Photos of other people. Only edit people who agreed to it. Photos of minors are declined.
Continue the workflow
Related tools for the next edit
Seedance 2.5 explained
ByteDance's video model: keep a clip's motion, change the look.
Genjutsu-style AI video
Turn your image into the look of a moving clip.
Change product background, keep the product
Original product pixels, new AI background.
Krea 3 guide
Krea 3 release status, what's confirmed, and how to try AI edits now.
Private by default
Your images and results are never public. No gallery, no indexable links.
Auto-delete in 24h
Anonymous uploads and results are wiped within 24 hours, automatically.
Never trained on
We don't use your uploads to train models. Not opt-out — it just doesn't happen.
Delete instantly
One click removes the source and every result. Download links die with it.
Simple pricing
A free first result. Credits when you need more.
5 credits per successful standard 1K image — including background changes with "Keep subject pixels exactly". Two more takes cost up to 10 credits. Failed images are refunded. One-time packs need no account to pay; new accounts get 5 free credits once.
Compare all credit plans and usage costs
Loading prices…
Subscriptions renew monthly and add credits after each successful payment. Packs are one-time purchases, available with or without a subscription. Unused credits carry over and remain after cancellation. No automatic pack purchases.
Honest limits
What AI still gets wrong — and what we do about it
No “100% accurate” promises. Here's the truth, and the workaround.
Faces can drift at Creative strength. If it stops looking like you, drop to Balanced or Subtle — Face lock works hardest there.
Hands and fingers remain the hardest thing in AI imaging. Keep them out of the change region when you can.
Tiny text and fine logos may blur or change. Turn on Text & Logo lock and compare them against the original before publishing.
Complex textures like lace, fur or micro-patterns can simplify. Subtle strength preserves them best.
Big scene swaps can nudge edges of the composition. Lock Composition and expect minor boundary shifts, not layout changes.
Failed generations happen. Timeouts and moderation blocks are labeled clearly and never consume your free quota.
FAQ
Questions, answered straight
Google's image generation and editing model with the official model code gemini-nano-banana-2.1. Google's API docs call it an update to Nano Banana 2 (Gemini 3.1 Flash Image) and the more efficient counterpart to Nano Banana Pro.
Not yet. The editor here currently runs Nano Banana 2 (Gemini 3.1 Flash Image). We'll switch this page to 2.1 when it's available to us and update this answer.
According to Google: better visual quality and realism at 1K, 2K and 4K, fixed tiling artifacts in very wide or tall images, better text rendering and infographic layouts, up to 14 reference images, and adjustable thinking levels.
Google offers it in the Gemini app and in AI Mode in Google Search (look for the banana icon). Developers can call gemini-nano-banana-2.1 through the Gemini API.
Say what to change and what to keep: 'turn this into a 1970s film still, keep the face and pose', 'make a product infographic with three short labels, keep the jar unchanged', 'redesign this room in Japandi style, keep walls and windows'. The same prompts work in the editor on this page.
For background changes, tick Keep subject pixels exactly: the original product or person is cut out in your browser and pasted back pixel-for-pixel, and the model paints only the background.
Hold Frame adds no visible watermark. Google embeds its invisible SynthID mark in images from its models.
No. Uploads are processed to make the result and are not used to train models.
Task-specific tools
Start from your exact job
Each tool pre-configures the right locks and presets for the task.
Sketch to Image AI
Line art to realistic or stylized renders. Outline & composition locked.
Change Background, Keep Person
New scene behind you, with your face kept as the reference. Face, body & pose locked.
Product Image AI
Studio shots and ad variants. Original product reference, new settings. Review labels and details.
Photo to Anime AI
Anime and manga styles with your face and pose intact.
Interior Design AI
Restyle a room without moving a wall. Layout, doors & windows locked.
Image Variation Generator
Four takes on the same idea. Subject & composition preserved.
AI Headshot Generator
Casual selfie to LinkedIn-ready headshot. Your face, locked.
Virtual Staging AI
Stage empty listings in seconds. Walls & windows never move.
Image Variation Prompts
Copy controlled prompts for lighting, palette, season, background and style.
Photo Variation Examples
Recreate practical before-and-after directions without losing the source.
AI Product Photography
Ecommerce and campaign photo workflows built from one real product.
Product Background Generator
Replace only the set while product, logo and packaging stay consistent.