AI news · October 9, 2026
Google's Nano Banana 2.1 supports 4K output and 14 reference images
Nano Banana 2.1 adds 1K, 2K, and 4K output, up to 14 reference images, search grounding, and three thinking levels, so image apps can test richer edits without moving to Pro.
- maximum output
- 4K
- reference images
- 14
- thinking levels
- 3
Google's model card and API docs list Nano Banana 2.1 as a Gemini 3 image model updated in October. It accepts text, images, video, and PDF input, returns images and text, supports 1K, 2K, and 4K output, and can fuse up to 14 reference images. The docs also list web and image search grounding and minimal, medium, or high thinking levels.
Google positions it as the faster, cheaper counterpart to Gemini 3 Pro Image. The model card's preference scores are higher than the older Nano Banana 2 and Pro on several listed tests, but those are Google's evaluations. It is available through Gemini, AI Studio, the Gemini API, Search AI Mode, Flow, and Stitch; small text and long paragraphs remain weak spots.
What you can do with it
Try model `gemini-nano-banana-2.1` in Google AI Studio with one product photo and two or three reference images. Compare a 1K draft with a 4K export, then test small labels and long text separately because Google lists those as weak spots.
Our take
The practical win is reference control, not another leaderboard number. Fourteen images make product or character consistency easier to test, while the documented text limits mean a maker still needs a separate layout pass for posters, menus, and dense diagrams.
Links Try it API docs Model card
Source: TLDR AI ↗ — Made With Models writes the brief; the reporting is theirs.