Google and Apple Expand AI Ties, Launch Nano Banana 2 Image Model Powered by Siri’s

Google’s new “Nano Banana 2” model—officially Gemini 3 Flash Image—marks the first public output of a deeper AI partnership between Apple and Google, according to a report from 富途牛牛. The collaboration routes Siri’s on‑device computational engine into Google’s server‑side image‑generation pipeline, letting the voice assistant’s low‑latency processing power act as a distributed compute node for the model’s heavy‑weight inference tasks. By offloading part of the workload to Apple’s silicon, Google claims the model can deliver “faster generation with significantly higher fidelity,” a boost that shows up in the demo videos posted on the Gemini app’s feed (Arjun, March 3).

Nano Banana 2 expands beyond the classic text‑to‑image workflow that made its predecessor a viral hit. As Arjun notes, the model now supports three distinct pipelines: pure text‑to‑image, image‑plus‑text editing, and multi‑image composition. The latter lets users feed a sketch and a screenshot and receive a polished UI mock‑up that blends both inputs—a feature that could streamline front‑end design and game‑asset prototyping. CNET’s coverage confirms that the new capabilities are already available to all Gemini users, who can experiment with the model directly in the app without needing a separate subscription (CNET).

The technical novelty lies in how Siri’s on‑device neural engine is repurposed as a “hosting server” for the AI workload. The 富途牛牛 report explains that Apple’s voice assistant, traditionally sandboxed to protect user privacy, now runs a lightweight inference client that streams model fragments to Google’s cloud back‑end. This hybrid architecture reduces round‑trip latency and cuts the cost of pure cloud compute, a win for both companies as they vie for AI‑first developer ecosystems. TechCrunch’s brief on the launch highlights the cost angle, noting that the model’s efficiency could translate into lower pricing tiers for Google’s AI‑plus and Ultra plans, though the outlet does not disclose exact numbers.

From a strategic perspective, the joint effort signals a shift from the “siloed” AI experiments that Apple and Google kept separate in previous years. By pooling Siri’s edge‑compute capabilities with Google’s generative‑AI expertise, the partners aim to create a differentiated service that leverages Apple’s hardware ecosystem while expanding Google’s foothold in the creator‑tool market. The move also positions both firms against rivals such as OpenAI and Adobe, which are rolling out their own image‑generation suites. While neither company has released a formal valuation of the partnership, the rapid rollout of Nano Banana 2 suggests they view the collaboration as a competitive imperative rather than a side project.

Early adopters are already testing the model’s limits. Arjun shares a handful of “must‑try” prompts that showcase the model’s ability to render complex spatial relationships and apply style transfers across multiple inputs. Users on the paid AI‑Plus tier report that the model can generate high‑resolution assets in seconds, a speed boost that TechCrunch attributes to the combined Siri‑Google compute stack. As the ecosystem matures, developers may see deeper integration of Nano Banana 2 into iOS shortcuts and Google Workspace, turning the once‑novel image generator into a staple of everyday productivity.