A timeline of Nano Banana models
Google's naming convention for Gemini's image generator has been confusing, so here is the chronology:
| Model | Released | Available on | Key changes |
|---|---|---|---|
| Nano Banana | Early 2025 | Gemini Free, AI Studio | First image generation in Gemini. Basic quality, 1024x1024 max |
| Nano Banana Pro | November 2025 | Gemini Advanced, AI Studio | Better detail, more consistent style adherence, higher resolution (up to 1536px) |
| Nano Banana 2 | January 2026 | Gemini Free, AI Studio | Replaced original Nano Banana on free tier. Better text rendering, fewer artifacts |
| Nano Banana 2 Pro | February 2026 | Gemini Advanced, AI Studio | Current top model. Up to 2048x2048, best quality, photo-realistic mode |
Each new version improved image quality — better hands, more accurate text in images, sharper details, stronger style control. But the watermarking system stayed the same throughout.
The watermark is identical across all versions
This is the part most people want to know. Regardless of which Nano Banana model generated your image, the visible watermark is:
- The same banana-shaped logo
- The same position (bottom-right corner, flush with the edge)
- The same alpha blending formula
- The same opacity level (~15-20%)
The only variable is size. Images up to approximately 1024px on the longest side get a 48x48 pixel watermark. Larger images get a 96x96 pixel version. Since newer models (especially Pro variants) tend to output at higher resolutions, you will see the 96px version more often with Pro and 2 Pro outputs.
But the shape and blending are the same. A removal tool that works on the original Nano Banana watermark works on Nano Banana 2 Pro too.
Free tier vs Gemini Advanced: watermark differences?
None. Google applies the same visible watermark to both free and paid outputs. There were rumors on Reddit in late 2025 that Gemini Advanced (the $20/month plan) would remove the watermark, similar to how some stock photo services work. That did not happen.
What the paid tier gives you: access to the Pro and 2 Pro models, which produce higher quality images at higher resolutions. The watermark is still there. Google's position is that all AI-generated images should carry a provenance signal — the visible logo for humans, SynthID for machines.
What actually changed between versions
Since the watermark is constant, here is what actually improved across model versions, based on testing:
Nano Banana (original)
The first model. Decent at landscapes and abstract art, struggled with text, hands, and fine details. Outputs maxed at 1024x1024. Fast generation (2-4 seconds). Available on the free tier.
Nano Banana Pro
A significant jump in quality. Text rendering improved but was still inconsistent. Hands went from "frequently wrong" to "usually correct." Resolution bumped to 1536px. Generation time increased slightly (3-6 seconds). Required Gemini Advanced subscription.
Nano Banana 2
Replaced the original on the free tier. Text in images became mostly readable. Fewer "uncanny valley" artifacts in faces. Same resolution ceiling as the original (1024px), but the 1024px outputs looked notably better than the original's. Google also improved the content filter, reducing false positives that blocked harmless prompts.
Nano Banana 2 Pro
Current flagship. Resolution up to 2048x2048. Photo-realistic mode that produces images hard to distinguish from photographs at a glance. Text rendering works reliably for short strings (5-6 words). Hands and fingers are consistently correct. Available through Gemini Advanced and AI Studio.
Removing the watermark with Banana Clean
Since all Nano Banana variants use the same watermark, Banana Clean handles all of them with the same approach: reverse alpha blending.
original = (watermarked - alpha * 255) / (1 - alpha)The extension ships with both mask sizes (48x48 and 96x96) and selects the right one based on the downloaded image dimensions. Whether you are generating images with the free Nano Banana 2 or the paid 2 Pro model, the extension intercepts the download and removes the watermark before the file reaches your downloads folder.
Processing takes under 100ms regardless of model version. The math is the same — only the mask dimensions differ, and applying a 96x96 mask takes negligibly longer than a 48x48 one.
Will future models change the watermark?
Possible but unlikely to be a major change. Google has kept the same watermark shape and blending approach through four model iterations over more than a year. The shape is tied to the Nano Banana brand, not to any specific model version.
If Google does change the watermark — new shape, different position, different opacity — the Banana Clean extension would need updated masks. Previous updates (like adding the 96px mask when Pro introduced higher-resolution outputs) were shipped within days of Google's changes.
What will not change: the watermarking approach itself. As long as Google uses alpha blending to composite a known, static watermark onto the image, reverse alpha blending will reverse it perfectly. The math does not care what the watermark looks like.
FAQ
Which model should I use for the best image quality?
Nano Banana 2 Pro, if you have Gemini Advanced. On the free tier, Nano Banana 2 is the only option and is still good for most use cases. Both support watermark removal via Banana Clean.
Can I tell which model generated an image after downloading it?
Not from the image file itself. The PNG/JPEG metadata does not include the model name. Within the Gemini interface, you can see which model is active, but the downloaded file does not carry that information.
Does the watermark size affect removal quality?
No. The reverse alpha blending formula works the same way for both 48x48 and 96x96 masks. The maximum per-pixel error is 1/255 per channel regardless of mask size. Both produce results that are visually indistinguishable from an unwatermarked original.
I see "Gemini Flash" mentioned in AI Studio. Does it have a different watermark?
Gemini Flash is a lightweight model optimized for speed and cost. When it generates images, the same Nano Banana watermark is applied. Banana Clean removes it the same way.