Google
Organizing the world's information
Start new thread

Nano Banana 2.1 - Google's image model with sharper design and mask editing

by
•
Nano Banana 2.1 is Google's latest image generation and editing model. Google says it outperforms earlier Nano Banana models across the board, with better visual design, mask based editing, stronger subject consistency and more natural looking images. It also improves text rendering and consistency across turns, with Search grounding at 1K, 2K and 4K. Try it in Google AI Studio or the Gemini app, or build with it in the Gemini API and Google Cloud.

Add a comment

Replies

Best
Hunter
📌

Hey PH community 👋

Excited to hunt Nano Banana 2.1 today.

Google just shipped an update to Nano Banana 2, and they say it outperforms their earlier Nano Banana models across the board. The gains they call out are better visual design, mask based editing, stronger subject consistency and images that look more natural. Google posted a short look at it on X .

Here's some examples from the community:

(this is me playing with Nano Banana 2.1 via Netlify AI Gateway)

Mask based editing is the part I'm most excited about. Pointing the model at one region of an image and changing only that is what makes it usable for real design work, instead of rerolling the whole frame until it lands.

Subject consistency is the other one I'd test first. Storyboards, ad sets and brand mascots all depend on the same person, product or character staying recognizable from one image to the next.

For builders, it's already live in the Gemini API with 1K, 2K and 4K output and Search grounding.

You can try it in Google AI Studio right now:

Give it a spin and let the team know what you create with Nano Banana 2.1 🚀

PS: Tagging and if you're here for any additional comments/ thoughts.

Sure, but have they fixed that thing where it just keeps giving you the same image over and over?

Hunter

 looks like they did!

every version of this launch leads with "more natural looking" as the headline improvement, and every time that's also the line that makes the provenance problem worse. SynthID is mentioned in passing in Google's own docs but not here in the launch pitch, and that's backwards for the thing that's explicitly optimizing against the artifacts people use to spot a fake. if "looks more natural" is the selling point, "here's how to tell it's synthetic anyway" should be in the same sentence, not a separate page someone has to go looking for

Hunter

 good call out but i think its a market wide problem, not just Google