Overview

• Google confirmed DeepMind built the model and integrated it into the Gemini app and website for free and paid users worldwide.
• All images created or edited in Gemini now carry a visible watermark and an invisible SynthID tag for provenance.
• New capabilities include combining multiple photos, multi‑step edits that preserve scene context, and transferring styles or textures between objects.
• The system was previously evaluated on LMArena as “Nano Banana,” where user comparisons favored it by Elo-style scoring.
• Business Insider’s tests found notably strong small edits and detail preservation, though the model sometimes faltered when merging different faces; Adobe says it is surfacing the model in Firefly and Express as investors weigh AI’s impact on traditional tools.