Guides / Models

Nano Banana 2.1 on NOLGIA: native 4K, four references, exact text and 8:1

Google released Nano Banana 2.1 on October 6, 2026, and it is in Create images now. Here is what changed, with the sources, and one run for each new feature: the exact prompt, the settings and the live price.

  • 7 min read
Native 4K4 referencesExact text8:1 wide

What you make, in order. Tap one to jump to its step.

Nano Banana 2.1 is Google's newest image model, an update to Nano Banana 2. Google released it and made it generally available on October 6, 2026. On NOLGIA it is in Create images now, from the Starter plan.

We ran each new feature once in Create images, on a phone, with three AI characters: Theo, Juno and Jean Phil. Every picture below is what that run returned, and the reel at the top is cut from the same runs.

The reel

The reel this guide is built from, in three parts. Each step below shows the same run with its prompt and settings.

The reel, part one: Jean Phil, Juno and Theo stacked in a human tower in one tall 1:8 picture, the title Nano Banana 2.1, Theo at a fruit stand asking what is new, and the four new features
Part one: the hook and what is new. The tower is one 1:8 image. Theo, the host, is an AI character: his lines are one Seedance 2.5 take, in his own voice.
The reel, part two: Nano Banana 2.1 picked in Create images, a 4K flea market where Jean Phil hides, four reference photos becoming one diner photo, and a letter board menu spelled exactly
Part two: native 4K, four references and exact text, captured in Create images on a phone. The narration is a clone of Theo's voice, made on NOLGIA.
The reel, part three: the 8:1 aspect ratio picked, a white stretch limousine across one ultra-wide picture, Theo in a print shop comparing the 4K price of Nano Banana 2 and 2.1, and Theo in a diner with the call to try it
Part three: 8:1, the price and the free guide.

What Google changed

Google lists these updates on its Gemini API model page for Nano Banana 2.1 (model id gemini-nano-banana-2.1), dated October 6, 2026:

  • Better quality and realism at 1K, 2K and 4K.
  • Cleaner wide and tall frames. Google fixed the tiling artifacts on the 1:4, 4:1, 1:8 and 8:1 ratios at 2K and 4K.
  • Better text. Enhanced text rendering and more accurate infographic layouts.
  • More references. Up to 14 reference images in one image through Google's API.
  • Grounding and thinking. Grounding with Google Web and Image Search, and thinking levels from minimal to high.

Google calls it the more efficient counterpart to Nano Banana Pro. The release date is in the Google Cloud release notes, and Google DeepMind's model page adds better visual design, subject consistency and legible text in images.

What that means on NOLGIA, checked on October 9, 2026. The live prices are in the cost block at the end of this guide.

Nano Banana 2.1Nano Banana 2Nano Banana Pro
2K and 4KRendered by the modelThrough NOLGIA's upscalerThrough NOLGIA's upscaler
Price at 2K and 4KLowest of the threeHigherHighest
Reference imagesUp to 411
Widest and tallest frames8:1 and 1:84:1 and 1:421:9 and 9:16

Pick Nano Banana 2.1 and set it to 4K

Open Create images, press the model and pick Nano Banana 2.1. Under Quality, pick 4K. The Generate button shows the price before you render.

The model list in Create images with Nano Banana 2.1 highlighted, showing its price, 1K and 4 references
Pick Nano Banana 2.1.
The Quality menu set to 4K (+5 cr)
Quality: 4K.
The Generate images button with the price of one 4K image
The price, shown before you render.
The prompt we used (9:16, 4K, Jean Phil added from Characters)
A 4K photograph of a packed Sunday flea market on a narrow Paris street, shot from a high balcony with a long lens: hundreds of shoppers between antique stalls, rugs, gilded mirrors, old cameras, crates of records, striped awnings, cafe tables and a brass band in one corner. Hidden very small in the crowd near the top of the frame, Jean Philanthrope in his brown houndstooth suit shadowboxing completely deadpan in front of a mirror stall, easy to miss. Natural afternoon light, crisp detail everywhere, every face and sign sharp, 35mm film colour. No text overlays.
A packed Paris flea market seen from a balcony: hundreds of shoppers between striped awnings, rugs, gilded mirrors and a brass band, with Jean Phil small among them
The 4K render, 3072 × 5504. Jean Phil is in there.
A crop of the same render: Jean Phil in his houndstooth suit shadowboxing in front of a gilded mirror stall
Found him. A crop of the same render at full resolution.

Tip:Draft at 1K. 1K is the cheapest size. Try the prompt at 1K, then render the one you keep at 4K.

Add up to four reference photos

Under the prompt, press Add reference image and upload up to four photos: faces, a product, a place. Each one gets a number, and the prompt points at it as @Image1 to @Image4. Say which photo is which person or thing, and what each one does.

The composer row: Nano Banana 2.1, Aura, Add reference image, Characters, Location, Brand kit and Product
Add reference image, next to Characters, Location, Brand kit and Product.
The upload box and References (4/4): Jean Phil, Juno, Theo and a RED LINE soda can, numbered Image 1 to Image 4
References (4/4). Image 1 is a reference photo by @jean_philanthrope. Juno and Theo are our own AI characters, and RED LINE is a made-up soda.
The prompt we used (4:5, 2K)
A flash photo of three friends squeezed into one red vinyl diner booth at night, like a photo booth portrait. @Image1 is Jean Phil: he sits completely deadpan in his brown houndstooth suit, holding a can of RED LINE cherry soda up like a trophy. @Image2 is Juno: in her red leather trench coat and oval sunglasses, she ignores the soda and sips a tiny white espresso. @Image3 is Theo: he laughs with his arms around both of them. The can is exactly the RED LINE can in @Image4, its wordmark sharp. Keep each face exactly like its photo. Hard direct on-camera flash, glossy skin, 35mm film grain, a neon sign glowing behind them. No other text.
Jean Phil holding up a RED LINE can like a trophy, Theo laughing with his arms around him and Juno sipping a tiny espresso, in a red diner booth under a neon sign
All four kept: three faces and the can, its wordmark readable. The prompt gave the neon sign no words, so the model made some up: give every sign its exact words.

Tip:Keep the four photos under 10 MB together. The reference images of one job can total 10 MB at most, and a bigger set fails before it renders (the credits come back). Save phone photos and PNGs as JPEGs about 1,000 px on the long side. Characters count too: their stored photos are sent as they are. Choose from assets holds one image at a time, so add the others as uploads.

Put the exact words in the prompt

Write the words exactly as they should appear, in order, and say so: *exactly these six lines in this order*. Separate the lines with a slash. This one needs no references.

The prompt box in Create images ending with the menu lines, Nano Banana 2.1 picked below it
The end of the prompt, typed.
The prompt we used (4:5, 2K)
A photograph of an old diner's black felt letter board menu hanging on a tiled wall, lit by a hard direct flash, white plastic letters, every word perfectly spelled, exactly these six lines in this order: JUNO'S DINER / ESPRESSO, SIPPED SLOWLY 4 / CREPE, ON FIRE, IGNORED 9 / OSTRICH SPECIAL, ASK INSIDE / SHADOWBOXING LESSONS 12 / NO PANIC SERVED HERE. On the counter below the board, a tiny white espresso cup and a crepe on fire on a plate, nobody reacting. 35mm film grain, saturated colour.
A black felt letter board on a tiled diner wall reading JUNO'S DINER, ESPRESSO, SIPPED SLOWLY 4, CREPE, ON FIRE, IGNORED 9, OSTRICH SPECIAL, ASK INSIDE, SHADOWBOXING LESSONS 12, NO PANIC SERVED HERE, with an espresso cup and a crepe on fire below
Six lines, spelled as typed, on the first run.
  • Keep each line short: a sign, a menu or a label reads better than a paragraph.
  • Type the capitals you want.
  • Read every word before you post. Google says text rendering improved, not that it never slips.

Go 8:1 wide or 1:8 tall

Under Aspect ratio, Nano Banana 2.1 adds 8:1 and 1:8, the widest and tallest frames any Nano Banana has on NOLGIA. Say what fills the length: what is at each end and what repeats in between.

The Aspect ratio grid with 8:1 selected and 1:8 on the last row
8:1 picked. 1:8 is on the last row.
8:1, 4K (Jean Phil and Juno added from Characters)
One continuous ultra-wide 8:1 photograph: the longest white stretch limousine in the world fills the whole width of the frame, parked on an empty desert highway at golden hour. At the far left end, Jean Philanthrope sits deadpan behind the wheel in his brown houndstooth suit, shadowboxing through the open driver's window. At the far right end, Juno leans out of the very last sunroof in her red leather trench coat and oval sunglasses, sipping a tiny white espresso. Dozens of identical tinted windows in between, a single tumbleweed rolling past the middle. Hard sun, long shadows, saturated colour, 35mm film grain. No text.
A white stretch limousine with dozens of windows across one ultra-wide desert picture, Jean Phil at the wheel at one end and Juno in the last window at the other, and below, the two ends cropped larger
One 8:1 picture, 11712 × 1408. The whole frame on top; below, its two ends, cropped from the same picture.
1:8, 4K (all three added from Characters)
One extremely tall vertical photograph in a 1:8 frame. A human tower of three people balanced on each other's shoulders in a narrow Paris courtyard at dusk, shot from far below with a wide lens and a hard direct on-camera flash. At the very top, Jean Philanthrope stands completely deadpan in his brown houndstooth suit, shadowboxing at the sky. In the middle, Juno in her red leather trench coat and oval sunglasses calmly sips a tiny white espresso, not holding on to anything. At the bottom, Theo in his brown leather blazer grins up at them, holding Juno's boots. Pigeons sit along the window ledges, a laundry line crosses the courtyard, and a concierge watches from a doorway, unbothered. 35mm film grain, saturated colour, deep blue sky. No text.
A human tower in a narrow Paris courtyard in one very tall picture: Jean Phil on top shadowboxing, Juno in the middle sipping an espresso, Theo at the bottom holding her boots; on the right, the top and the bottom cropped larger
One 1:8 picture, 1408 × 11712. The whole frame on the left; on the right, its top and its bottom, cropped from the same picture.

Tip:Where the long frames fit. 8:1 suits banners, website headers and panoramas. 1:8 suits a tall scroll or a story that reads from top to bottom.

What 2.1 does not do on NOLGIA

  • One image per request.
  • No mask editing. Google's DeepMind page mentions mask-based editing; Create images does not offer it for Nano Banana 2.1.
  • No thinking level or search grounding controls.
  • Four reference images per prompt, not the fourteen of Google's API.

What it costs

Every image in this guide is one Nano Banana 2.1 run at 2K or 4K. Nano Banana 2.1 works from the Starter plan, and the Generate button always shows the exact price first. Here are the live prices next to Nano Banana 2 and Nano Banana Pro.

  • Nano Banana 2.1Every plan

    Per image

    1K 4 credits · 2K 5 credits · 4K 9 credits

  • Nano Banana 2Every plan

    Per image

    Native 4 credits · 2K 17 credits · 4K 54 credits

  • Nano Banana ProEvery plan

    Per image

    Native 8 credits · 2K 21 credits · 4K 58 credits

Read live from the catalog when this page loads. Every price is shown before you generate. See every rate.

Questions and answers

Is Nano Banana 2.1 a new model or an update?
An update to Nano Banana 2 (Gemini 3.1 Flash Image), released and generally available on October 6, 2026. Google calls it the more efficient counterpart to Nano Banana Pro.
Is it cheaper than Nano Banana 2 on NOLGIA?
At 1K they cost the same. At 2K and 4K Nano Banana 2.1 costs much less, because it renders those sizes itself while Nano Banana 2 goes through NOLGIA's upscaler. The cost block in this guide shows the live prices.
How many reference images can I use?
Up to four per prompt on NOLGIA, as uploads or characters, 10 MB together at most. Google's API accepts up to 14. Nano Banana 2 and Nano Banana Pro take one on NOLGIA.
Which plan do I need?
Starter or above.
Who are the people in the video?
AI characters. Theo and Juno are NOLGIA originals. Jean Phil is our recreation of the AI influencer @jean_philanthrope from his reference photos, and NOLGIA is not connected to his account. RED LINE is a made-up soda.

Try Nano Banana 2.1

Open Create images

Filmmaking

Keep characters and locations consistent across shots

A saved character, a saved location and one repeated sentence per prop: what actually held across five shots of one film, measured object by object and by color, and the two gaps that did not.

· 11 min read

Product images

One phone photo to a full shop set

Turn one clear phone photo into a studio hero, a seasonal version, a product page banner and a short clip, with the preset for each step and what the set costs.

· 6 min read