Google has replaced Nano Banana 2, its everyday image model since February, with Nano Banana 2.1. It makes and edits pictures at about half the old price, and it is rolling out in the Gemini app, AI Mode in Search, Google Ads and the tools developers use to build their own products.
For developers, a standard 1K image now costs 3.36 cents instead of 6.70, and a 4K image 7.56 cents instead of 15.10. Most of what the model can do was already there: it could mix up to 14 reference images and look things up on Google Search before. What is different is that the pictures look better, the text inside them is more reliable, and the price is lower. Anyone who built a tool on the old model has until October 29 to move, because Google is switching it off.
Nano Banana Pro, the more expensive model, often still gives noticeably better pictures, at 13.40 cents for a 1K image, about four times the new price. The sensible split is the cheap model for trying ideas and Pro for the picture that matters most.
Stock photography showed where the value goes
A cheaper picture does not make pictures less valuable. It moves the value somewhere else, and stock photography is the clearest case. Before microstock sites, customers paid $600 or $1,200 for a licence. Microstock sold the same kind of picture for about a dollar, and the market changed shape.
The photographers who did well were the ones who treated it as volume. The best known of them sells more than 10,000 licences on a typical day, at about $1 each, from an archive of more than 600,000 photos. Most stock photographers earn around 30 cents a licence, and about 10% of a portfolio brings in 90% of its sales. The sites now hold hundreds of millions of photos, so nobody is short of pictures. What is short is the ability to pick the right one.
What a small shop does with 5-cent pictures
Take a shop with 200 products that wants five different scenes for each, such as a kitchen counter, a picnic table and a gift box. That is 1,000 images. At the 2K price of 5.04 cents, it comes to about $50. Suppose nine out of ten pictures are not good enough. The shop generates 10,000 images to end up with 1,000, and the bill is about $500.
The model can keep up to ten objects the same across a set of reference pictures, so the shop starts by giving it real photos of each product. The money is no longer the problem. The time is: if someone looks at each of the 1,000 pictures for ten seconds, that is nearly three hours, and in that time they have to see that the jar is the right shape, the colour is the colour customers will receive, and the label is spelled correctly. A customer who gets a product that does not look like the picture will return it, and that cost is much bigger than the 5 cents.
The same applies to anything with facts in it. Nano Banana 2.1 is better at infographics, but on the benchmark that measures whether the information in them is right, it scores 0.52 with thinking turned on and 0.33 with it off. The settings are minimal, medium and high thinking, and the higher ones make better pictures. A chart for a customer or a school newsletter still needs someone to check the numbers.
So the question for a team is not whether it can afford more pictures. It is who looks at the thousand pictures, and who has the authority to reject most of them.