Google has launched Nano Banana 2, the latest iteration of its image generation model, which the company claims is imbued with “advanced world knowledge” and “precision text rendering and translation.” This new version is described as a “best of both worlds” mashup of the original model and the advanced Nano Banana Pro, released last November. Running on Google's Gemini 3.1-Flash model, Nano Banana 2 promises faster image generation than ever before.

A key strength of Nano Banana Pro, its ability to handle text, has been further refined in Nano Banana 2. The model can now leverage the real-world knowledge base of Gemini, Google's flagship LLM, to pull real-time information and images directly from web search for more accurate rendering of specific subjects, making it ideal for infographics and data visualizations. Creative control has also been enhanced, allowing the generation of up to five characters and 14 objects in a single workflow, with improved adherence to user prompts and better lighting and texture effects.

While many features are rehashed from previous models, the primary selling point of Nano Banana 2 is its speed, thanks to the Flash model. It significantly reduces image generation time and supports resolutions ranging from 512px to 4K. Nano Banana 2 is set to replace older models across Google's suite of applications, including the Gemini app, Google Search, Flow, AI Studio + API for developers, Google Cloud, and Google Ads.

👉 Read the full article on Gizmodo

Huxe is an AI-powered audio assistant designed to deliver personalized daily audio summaries, effectively creating AI-generated “podcasts.” The application analyzes user data, such as email communications and calendar schedules, to tailor content to individual interests and daily routines. This approach aims to combat information overload by providing users with a hands-free, continuous listening experience.

Inspired by Google's NotebookLM and developed by former Google engineers, Huxe allows users to interact with multiple AI hosts, ask follow-up questions, and delve deeper into specific topics. This interactive capability encourages users to transition from passive listening to more active information engagement.

Currently available for free on iOS and Android, Huxe seeks to transform a user's digital life into a 24/7 personalized radio station. However, some feedback suggests that its execution can sometimes feel more like text-to-speech than a natural podcast, indicating areas for further refinement.

👉 Read the full article on Wired

Sammy Azdoufal, a software engineer and AI strategist, inadvertently discovered a security vulnerability that granted him access to nearly 7,000 camera-enabled DJI Romo robot vacuums across 24 countries. This occurred while he was developing a remote-control app for his own DJI robot vacuum, using an AI coding assistant, specifically Claude Code, to reverse-engineer its communication protocol.

During this process, Azdoufal found a backend security bug where the credentials for his own device also provided access to live camera feeds, microphone audio, maps, and status data from thousands of other robot vacuums. Instead of exploiting the flaw, he reported it to The Verge, which then contacted DJI, leading to a swift resolution of the issue.

The incident underscores long-standing cybersecurity warnings about internet-connected smart home devices and highlights how AI-powered coding tools can inadvertently amplify such vulnerabilities. Experts have consistently cautioned that these devices present attractive targets for hackers, and this event serves as a stark reminder of those risks.

👉 Read the full article on Wired