A new entry in the Vertex AI model catalog was spotted: model:gemini-3.1-flash-image.
It looks like the rumors were true—this is the official identity of Nano Banana 2. While everyone was waiting for a Pro update, Google seems to be doubling down on the "Flash" tier for high-volume production.
Here’s the breakdown of what this means for production:
The Pro vs. Flash: Based on early internal samples, the quality is surprisingly close to Nano Banana Pro. In some dense compositions, the Flash model actually seems to handle spatial logic better than the flagship.
Put them to the test with the same prompt. The left is generated by Nano Banana 2/Gemini 3.1 flash image, and the right is Nano Banana Pro called via AtlasCloud.ai. To my eyes, the gap is almost invisible. You could see it from here:
https://www.reddit.com/r/nanobanana2pro/comments/1rea91s/nano_banana_2_is_realgemini_31_flash_image_just/
Built for Scale: The naming convention confirms this isn’t a Pro replacement, but a high-speed, low-cost alternative.
Feature Parity: It’s inheriting all the features from the Nano Banana series:
Multi-subject reference
High-fidelity style transfer.
Precise semantic following.
This is clearly aimed at high-frequency pipelines—think bulk UGC ad creation, or generating consistent frames for video models like Kling 3.0 or Seedance 2.0. If the pricing is as low as the previous Flash models, this might be the most important release for H1 2026.
Nano Banana 2 is real！Gemini 3.1 Flash Image just appeared in Vertex AI Catalog | Heykuki News