Nano Banana 2.1
Masked Editing, Consistent Subjects

Nano Banana 2.1 is the updated release of Google's Nano Banana image generation and editing model. Compared with earlier versions, it leans hard into controllability: mask-based local editing, stronger subject consistency across multiple outputs, multi-reference composition, and more legible in-image text. Google also distributed it inside Flow and AI Mode in Search, so it reaches a far wider audience than a developer-only preview would.

Think of it as different jobs. Nano Banana 2 Lite optimizes for speed and cost per image — it is the one you reach for when you need fifty drafts cheaply. Nano Banana 2.1 optimizes for control and fidelity: masked editing, holding a subject steady across a set, and composing several references into one frame. If you are iterating on concepts, Lite is fine. If you are editing an approved asset or building a consistent campaign, use 2.1.

You draw or select a region of the image — a mask — and only those pixels get regenerated from your instruction. The rest of the frame is preserved. That matters because the usual failure mode of image models is that asking for one small change rerenders everything, and the product, the face, or the brand color shifts slightly. Nano Banana 2.1 masked editing avoids that by scoping the generation to the area you chose.

Yes — subject consistency is one of the two features it is built around. Supply a clear reference image and reuse it across your requests, and the model carries facial features, hairstyle, clothing details, and packaging design into each new scene. It is not a guarantee of pixel-identical output, so for high-stakes work generate a few candidates per scene and keep the closest match as your new reference going forward.

Google prices the Nano Banana models per generated image, and the exact rate depends on which surface you use — Flow, the Gemini API, or Search — and on your tier, so check Google's current pricing page before you budget a large batch. On VO3 you do not manage a separate image bill: generation runs on the same credit balance you use for video, and you can start on the free tier to see the output quality before you pay anything.

That is exactly what VO3 is for. Upload your Nano Banana 2.1 still as the first frame on the create page, add a short motion prompt, and generate with Veo 3, Veo 3 Fast, Kling, Hailuo, or Seedance. Because your subject is already consistent across the image set, the resulting clips cut together as one sequence rather than five unrelated shots. The Veo models also generate synchronized audio, so a product spot can come out of a single pass.

Google exposes the model through Flow and AI Mode in Search, both of which have free usage within limits. On VO3, new accounts get starter credits with no card required, which is enough to generate a handful of images and animate one or two of them end to end. Most people use the free pass to confirm that masked editing holds their brand asset correctly, then upgrade when they start producing in volume.

Images and videos you generate on a paid VO3 plan are yours to use in ads, listings, and client deliverables. Two practical cautions regardless of platform: do not upload reference photos of real people without their permission, and do not mask-edit trademarked packaging that is not yours. The subject-consistency feature makes both of those mistakes much easier to make at scale, so set the rule with your team before you start a batch.