Google's Gemini image models, online and without watermark. Nano Banana Pro for print-grade detail, Nano Banana 2 for Flash-speed iterations, both with an edit mode. 3 credits per image on Aura AI.
Nano Banana is the name Google gives to the image generation models inside Gemini. The nickname started as a codename for Gemini 2.5 Flash Image in August 2025 and stuck because the model became the reference for identity-preserving edits: hand it a photo of a person or a product and get back a new scene with the same face, the same packaging, the same details. Google DeepMind has since shipped two successors. Nano Banana Pro, announced on November 20, 2025, is Gemini 3 Pro Image, the quality-first model with 2K and 4K output upstream, accurate multilingual text and Google Search grounding. Nano Banana 2, announced on February 26, 2026, is Gemini 3.1 Flash Image, which brings the Pro model's reasoning, world knowledge and consistency to a much faster tier.
What sets the family apart is the combination. Both models keep up to 14 reference objects and up to five people consistent across a series of images, so a character sheet or a product catalog stays coherent from frame to frame. Both render legible text in several languages, which makes them practical for mockups, posters and social graphics rather than only art. Both can pull real-world context from web search, so a specific landmark or dish is grounded rather than guessed. Every output carries Google's invisible SynthID watermark and C2PA content credentials upstream, which identify the picture as AI generated without altering how it looks.
On Aura AI, Nano Banana Pro and Nano Banana 2 run in the text-to-image tool at a flat 3 credits per image, in five aspect ratios. Both also power the image editor, where Nano Banana Edit costs 2 credits and Nano Banana 2 Edit costs 3. Output has no visible watermark, and the same credits work on GPT Image 2, FLUX 2, Seedream 5.0 and every other model in the studio.
Six reasons Google's image models became the default for consistent, text-heavy visuals
Both models preserve the resemblance of up to five people across a sequence of images. Generate a mascot or a comic character once, then place it in new scenes without the face drifting.
Both models accept up to 14 reference objects in one workflow upstream. Combine a product photo, a logo and a location shot and the model fuses them into one coherent composition.
Headlines, labels and captions come out readable, in multiple languages. Nano Banana 2 can even translate the text inside an existing image while keeping the layout.
Nano Banana Pro offers 2K and 4K output upstream, and Nano Banana 2 spans 512px to 4K. That is why Pro is Google's pick for print, packaging and large-format assets.
Google connects the models to real-time information and images from search. Ask for a specific monument or recipe and the result reflects what it actually looks like.
Nano Banana 2 is built on the Gemini Flash tier, so it returns Pro-level quality in a fraction of the time. Use it when you iterate on a concept many times before committing to a final.
Four steps from prompt to a finished image, or from a photo to a clean edit
Flat cost per image, no resolution or aspect ratio multiplier
| Variant | Mode | Credits per image | Best for |
|---|---|---|---|
| Nano Banana Pro | Text to image | 3 credits | Print-grade detail, dense text, product and brand assets |
| Nano Banana 2 | Text to image | 3 credits | Fast iterations, storyboards, social content at volume |
| Nano Banana Edit | Image edit | 2 credits | Background swaps, object removal, restyling your photos |
| Nano Banana 2 Edit | Image edit | 3 credits | Multi-step edits and text changes with the newest model |
How Google's image models compare to the other flagship generators on Aura AI
| Feature | Nano Banana | GPT Image 2 | FLUX 2 | Seedream 5.0 |
|---|---|---|---|---|
| Maker | Google DeepMind | OpenAI | Black Forest Labs | ByteDance |
| Latest release | Feb 2026 (Nano Banana 2) | Apr 2026 | Dec 2025 (FLUX 2 Max) | Jul 2026 (5.0 Pro) |
| Credits, text to image | 3 | 4 | 1 to 3 | 3 to 4 |
| Credits, image edit | 2 to 3 | 4 | 2 to 3 | 2 to 4 |
| Reference images (upstream) | Up to 14 | Multi-image | Up to 10 (Max) | Up to 14 (4.5 Edit) |
| Max resolution (upstream) | 4K | 3840 x 2160 | 4 megapixels | 4K |
| Text rendering | ✓ Multilingual | ✓ Multilingual, CJK | ✓ Yes | ✓ Chinese and English |
| Plan | All plans | All plans | All plans | All plans |
Where consistency, references and clean text pay off
Upload a single packshot in the editor and ask for new settings, angles and lighting. The bottle, label and logo stay identical while the scene around them changes.
Nano Banana Pro renders headlines and body copy legibly, so a launch poster or a quote card comes out in one pass instead of being assembled in a design tool afterwards.
Design a character once, then reuse it across a storyboard, a children's book or a mascot campaign. Both models hold face, outfit and proportions across new poses.
Grounded in world knowledge and good with structured layouts, the models produce diagrams, app screens and explainer visuals that make sense rather than only look plausible.
Common questions about Nano Banana Pro and Nano Banana 2 on Aura AI
Google's Nano Banana Pro and Nano Banana 2, 3 credits per image, edit mode included, no watermark. Available on every plan, worldwide.