Google adds 'Nano Banana' in Gemini
Google rolled out 'Nano Banana' inside Gemini as a high‑end AI photo editor and shared a set of 12 example edits showcasing generative photo capabilities. The feature is another example of Big Tech embedding studio‑grade content tools into platform LLM products. (x.com)
Google replaced its earlier Nano Banana models with “Nano Banana 2” (branded as Gemini 3.1 Flash Image) in a staged rollout that Google recorded on February 26, 2026 in the Workspace release notes. (workspaceupdates.googleblog.com) Nano Banana / Nano Banana Pro variants are exposed not only in the Gemini app but also through Google’s AI Studio and the Gemini/Vertex APIs, with Google noting quick API setup and integration paths for teams. (aistudio.google.com) (picsman.ai) Feature-level specs include native 1K output with built‑in upscaling to 2K and 4K, and the Workspace notes state Nano Banana 2 preserves subject consistency for up to five characters and fidelity for up to ten objects in a single workflow. (aistudio.google.com) (workspaceupdates.googleblog.com) Google adds provenance controls: images from Gemini 3 Pro Image include an invisible SynthID watermark to mark AI‑generated content, and Google’s product post explicitly flags generative image editing as experimental while showing curated example edits. (aistudio.google.com) (blog.google) Access tiers and runtime behavior differ by model: reviewers observed that Nano Banana Pro (Gemini 3 Pro Image) gives paid subscribers priority access while free users fall back to the original Nano Banana model after hitting usage caps. (pcmag.com) Concrete interview- and portfolio‑focused prompts inspired by the rollout: implement a prototype inference pipeline that preserves a subject identity across five edited scenes and upscales outputs to 4K using an external upscaler (aligns to Nano Banana 2’s 5-character / 4K claims); instrument metrics for per‑edit cost and watermark cover rate; and wire an API wrapper that falls back to a lightweight model after a quota is reached to mirror Google’s tiered behavior. (workspaceupdates.googleblog.com) (aistudio.google.com)
