Google has updated its Nano Banana image generation and editing model. The new version offers improved handling of text, characters, and multiple images, and is noticeably cheaper than its predecessor, according to infohub.kz.
The new Nano Banana 2.1 runs on Gemini 3.6 Flash and can accept up to 14 images at once, helping the model better understand what needs to be created. It can also process a larger amount of information in a single request, and characters and objects now retain their appearance more consistently when new images are generated.
Nano Banana 2.1 has a better memory of how characters and objects look, so they do not change from image to image. For example, if you create several scenes featuring the same four characters, the model tries to keep their appearance consistent. The same applies to objects.
The model can also search Google for information if it is needed to create an image. For instance, you can ask it to make a picture featuring a specific place, person, or object, and it can use data from search.
Separately, Google has cut the cost of generation. A 1K image now costs $0.0336 (about 15 tenge) instead of $0.0670 (about 30 tenge), while a 4K image will cost $0.0756 (about 34 tenge) instead of $0.1510 (about 67 tenge). For comparison, the flagship Nano Banana Pro costs $0.1340 (about 60 tenge) per 1K image.
That said, the model is not yet perfect. It can still get confused about the placement of objects, and small text in 1K images sometimes comes out blurry.