Google launches Nano‑Banana 2
Google released Nano‑Banana 2, touting improved subject consistency and sub‑second 4K image synthesis — and clarified the difference between 'Google‑Agent' (user‑triggered AI) and 'Googlebot' (web crawling) to mark a technical boundary for AI integrations. That formal separation should matter for developers who design AI features vs. search optimizations. (marktechpost.com)
Google published the Nano‑Banana 2 announcement (branded as “Gemini 3.1 Flash Image”) on February 26, 2026 in a company blog post describing the model as the successor that combines Nano Banana Pro capabilities with Gemini Flash speed. (blog.google) Multiple technical writeups and model pages report Nano‑Banana 2 as a ~1.8 billion‑parameter image model that uses techniques named Dynamic Quantization‑Aware Training and Grouped‑Query Attention to target sub‑500 ms inference and claim up to 30 fps at 512 px. (ubos.tech) Google’s Workspace rollout notes that Nano‑Banana 2 replaces earlier Nano Banana variants across the Fast, Thinking and Pro options for Workspace customers and signed‑in personal accounts, and the model maintains subject consistency for up to five characters and fidelity for up to ten objects in a single workflow. (workspaceupdates.googleblog.com) The company also said Nano‑Banana 2 includes improved precision text rendering and image localization features and that Google is extending SynthID work alongside C2PA Content Credentials to label AI‑generated images for provenance. (workspaceupdates.googleblog.com) Separately, Google added a new user agent string, “Google‑Agent,” to its user‑triggered fetchers documentation on March 20, 2026 and provided example HTTP user‑agent strings and IP ranges to distinguish agent‑initiated web visits from background crawlers such as Googlebot. (searchengineland.com) Google’s developer documentation classifies Google‑Agent under user‑triggered fetchers, notes those fetchers may ignore robots.txt rules, and recommends verification via reverse DNS or the published JSON IP lists (including user‑triggered‑agents.json) for sites wanting to identify or allow agent traffic. (developers.google.com)
