LMArena is a community-driven AI benchmarking platform where anonymous model outputs are compared via real user votes. Recently, a mysterious image editing model called Nano-Banana—suspected to be a Google prototype—has emerged in Arena battles, notable for its natural language editing, consistency, and near real-time responses.
LMArena is a research-driven, public web platform—built by UC Berkeley’s SkyLab—that lets users evaluate AI models (text and vision) in real time via anonymous head-to-head battles. With over 3.5 million human preference votes and 400+ models tested, it shapes an evolving leaderboard based on genuine user feedback.
Nano-Banana is a stealth image generation model currently appearing within LMArena’s Image Edit / Battle Mode, and it's drawing significant attention for its exceptional editing and prompt fidelity.
Likely Google-related: The name and early usage patterns suggest an internal Google model, possibly tied to “nano” class models like Gemini Nano.
Limited exposure: Nano-Banana appears randomly—not selectable—during Arena battles. Access is inconsistent and exclusive for now.
Community buzz: Users report mind-blowing outputs—sharp edits, context-aware transformations, and fully preserved character coherence.
“It keeps diagrams, lighting, scene logic—this isn’t just another experiment.”
Can I directly choose Nano-Banana?
No. It's randomly featured in the Image Battle Arena—there’s no manual selection.
Is Nano-Banana officially confirmed by Google?
Not yet. Community hints suggest a possible Google link, but there's no official confirmation.
Can I use Nano-Banana outputs commercially?
Usage rights are unclear—especially as Nano-Banana is not officially released. For serious use, it's not recommended to rely on private or anonymous AI outputs.
Share your thoughts about this page. All fields marked with * are required.