Nano Banana 2.1: Google's fast image model, at half the cost per image
1 month ago
Google has released Nano Banana 2.1, its new fast model for making and editing images. It became generally available in the Gemini API on October 6th, as an update to Nano Banana 2.
Ask about this presentation
Chapters
0:00Nano Banana 2.1 is generally available
0:14The efficient counterpart to Nano Banana Pro
0:22Google reports better images, characters and text
0:31Panoramic images lose their tiling artifacts
0:44Google's raters preferred 2.1 in every category
0:57What you can do
0:58A new medium thinking level, now the default
1:10Mix up to 14 reference images
1:20It can check Google Search before it draws
1:31Cost and access
1:33Each image costs half as much
1:45Paid tier only in the Gemini API
1:58Also in the Gemini app, Search, Ads, Flow and Stitch
2:07Before you switch
2:08Nano Banana 2 shuts down on October 29, 2026
2:19The 512-pixel size isn't supported on 2.1
2:34What Google says 2.1 still gets wrong
2:46Every image carries a SynthID watermark
Show transcriptHide transcript
Google has released Nano Banana 2.1, its new fast model for making and editing images. It became generally available in the Gemini API on October 6th, as an update to Nano Banana 2.
Google pitches it as the efficient counterpart to Nano Banana Pro, its premium image model. It keeps Flash level speed and cost.
Google says the images look more realistic, characters stay the same from one edit to the next, and text in infographics comes out more accurately.
Nano Banana 2 added very wide and very tall images, up to eight to one. At 2K and 4K those could show tiling artifacts, and Google says version 2.1 fixes them.
In Google's side by side tests, human raters preferred 2.1, with thinking on, over Nano Banana 2 and Pro in every category. Its widest lead was keeping several characters consistent.
What you can do.
You choose how much the model thinks before it draws. Version 2.1 adds a medium level between minimal and high, and makes it the default. Nano Banana 2 started at minimal.
You can hand it up to fourteen reference images and combine them into one picture. It keeps up to four characters looking like themselves, and up to ten objects true to the originals.
With grounding on, the model can check Google Search and Image Search, so a weather map shows today's data. For now, it can't use real photos of people from those results.
Cost and access.
Each image costs half what Nano Banana 2 charged. That's just over three cents at 1K, and about seven and a half at 4K. Input tokens cost three times as much, a dollar fifty per million.
In the Gemini API, 2.1 is for paid accounts. Google's pricing page lists the free tier as not available. The Batch API halves the price, and Google AI Studio lets you try it.
Outside the API, Google's model card lists 2.1 in the Gemini app, AI Mode in Google Search, Google Ads, Flow and Stitch.
Before you switch.
If you build on Nano Banana 2, you have just over three weeks. Google has deprecated it, will shut it down on October 29th, and tells developers to move to 2.1.
Images come out at 1K, 2K or 4K, with 1K the default. Nano Banana 2 also made small 512 pixel images, and 2.1 doesn't support that size, so check it before you switch.
Google's model card lists the weak spots. Small text can come out blurry, often at 1K. Characters don't always match the input images. And it sometimes mixes up left and right.
Every image it makes carries Google's SynthID watermark. Google's Image generation guide has the examples to start from. That's Nano Banana 2.1, Google's fast image model, at half the cost per image.

