Nano Banana 2 AI Image Editing Model

Gemini 3.1 Flash with image generation capabilities. High-efficiency image generation model with support for text rendering, reference images, search grounding, and thinking mode. The efficient counterpart to Gemini 3 Pro Image.

What Nano Banana 2 is good at

Nano Banana 2 competes across 7 categories in the Lumenfall Arena, where real users vote between AI-generated images in blind comparisons without knowing which model produced which result. Each ranking reflects genuine human preference based on 19+ competing models. Nano Banana 2 particularly excels at #1 Product, Branding & Commercial, #2 Text Rendering, #4 Portrait, #5 Portrait, and #7 Photorealism — ranking in the top 40% of all competing models in these areas.

Categories where this model excels

Nano Banana 2 performed particularly well in these areas

Product, Branding & Commercial

Text-to-Image · 19 models competing View leaderboard
Nano Banana 2 — Ranked #1 of 19
Selected other models in this challenge
No other competitor
available

Judged using these challenges:

Text Rendering

Text-to-Image · 21 models competing View leaderboard
Nano Banana 2 — Ranked #2 of 21
Selected other models in this challenge

Portrait

Text-to-Image · 19 models competing View leaderboard
Nano Banana 2 — Ranked #4 of 19
Selected other models in this challenge

Judged using these challenges:

Portrait

Image Editing · 14 models competing View leaderboard
Nano Banana 2 — Ranked #5 of 14
Selected other models in this challenge

Photorealism

Image Editing · 16 models competing View leaderboard
Nano Banana 2 — Ranked #7 of 16
Selected other models in this challenge
No other competitor
available

Other categories

Nano Banana 2 competes but doesn't rank as highly

Anime

Image Editing · 13 models competing View leaderboard
Nano Banana 2 — Ranked #6 of 13
Selected other models in this challenge

Judged using these challenges:

Photorealism

Text-to-Image · 22 models competing View leaderboard
Nano Banana 2 — Ranked #14 of 22
Selected other models in this challenge

Judged using these challenges: