Nano Banana 2.1 Image Generator and Editor

Nano Banana is Google’s name for its Gemini image models. Google announced Nano Banana 2 on February 26, 2026 as Gemini 3.1 Flash Image, aiming to bring Pro level features at Flash speed: up to five consistent characters and 14 objects in a workflow, better text rendering and output from 512 px to 4K.

On October 6, 2026 Google released version 2.1 and, in the Gemini API changelog, marked the older model as deprecated and asked developers to move to 2.1. The Decoder reported that 2.1 improves text rendering, takes up to 14 reference images and costs about half as much in the API, while noting that Nano Banana Pro often still looks better in practice.

So this page runs version 2.1 through fal.ai, not the older model. Google’s image docs say every generated image includes an invisible SynthID watermark, so expect it on these files too.

Credits: 8 an image at 1K and 11 at 2K, 1 to 4 images a run; there is no 1.5K step for this model. Size: about 1 or 4 megapixels. The model itself goes to 4K (fal lists 1K, 2K and 4K), but 2K is the largest size we offer.

References: up to 10 images. Google allows 14; here the limit is 10, the same as Seedream 5.0 Pro and Muse Image. Shapes: 10 of our 11 aspect ratios, from 21:9 to 9:16, without 9:21. Time: usually 15 to 40 seconds; in our own test on October 9, 2026 it made a packaging design in 15 seconds and printed "MOSS & CO" and "NET WT 4 OZ (113 g)" correctly.

You get a PNG per image. Credits come with the plans on pricing, and a run that fails returns its credits.

GPT Image 2.5 Flare and Sunburst (OpenAI) cost 10, 12 or 15 credits and take 4 references; choose them when many lines of small text must be exact. Seedream 5.0 Pro (ByteDance) costs 11 or 22 credits, takes 10 references and is strong on materials, dense layouts and text in many languages; it also has 9:21. Muse Image (Meta) makes a 2 credit draft at one size, about 2.5 megapixels.

Google’s model is our pick when you start from images rather than words: a sketch, last year’s pack, a product you want in a new setting. It is also the cheapest way to a 2K file here. Compare examples on Seedream 5.0, GPT Image 2 and Muse Image.

Say what stays and what changes, in that order: "keep the same bottle shape and fox logo; add a knit sweater pattern to the label and snow on the shoulders". Without the first half the model may redesign the product too.

Use the clearest reference you have: a flat, even photo of the package, the label facing the camera. For a sketch, say "keep the layout of the sketch" so the panels land where you drew them. Add a second reference for style, such as a photo of a scene you like.

Type the words to print again in the fields even if they are on the reference. Small type on a reference label can be redrawn wrong, so check the brand, product name and net quantity at full size.

Type the net quantity the way it must print, and the model copies it. On US food packages it goes in the bottom 30% of the front panel in bold type (21 CFR 101.7), in both US and metric units (15 U.S.C. 1453), as in "NET WT 24 OZ (680 g)" on the sauce jar among the examples.

Leave barcodes and Nutrition Facts out of the image and build them exactly with our free UPC barcode generator and nutrition label generator. For print, place the artwork on a real box from the dieline generator, or design on the dieline from the start with AI packaging design, where this model is in the menu too.