Summary
Highlights
Google's NotebookLM now features cinematic video overviews, allowing the AI to analyze source material and generate a 5-minute video with a suitable structure and visual style. Powered by Gemini 3 Pro, it can create programmatic animations for precise details, such as historical maps or algorithmic visualizations, using various AI models like Nano Banana and V3 for consistent and standard video generation. It also includes an automated self-critique loop for error correction and narrative adjustment.
NotebookLM has introduced presets for infographic visual styles, ranging from professional to creative. A new editing feature allows users to revise slide decks, correcting errors or making stylistic changes efficiently. Users can now also generate infographics directly from chat conversations, allowing for more targeted and relevant visual summaries based on specific discussions.
Gemini now integrates LIA 3, Google's new music generation model, allowing users to create short music tracks in various styles directly within Gemini. This expands Gemini's multimodal capabilities, though it's limited to 30 generations. The video demonstrates this with an East Coast rap song generation.
Google acquired Producer AI (formerly Rift Fusion), which now uses Leria 3 to generate full songs. A key feature is the ability to edit generated songs using natural language prompts, adjusting mood, instrumentation, and vocal focus. The platform can generate lyrics or use user-provided lyrics and offers creative tools like synths and drum machines. While powerful, some complex adjustments may not follow perfectly, and it's compared to Suno for different styles and results.
Manis is an AI agent that orchestrates multiple models to perform complex, multi-step tasks such as video analysis, image generation, and website creation. It can conduct research, analyze content, and compile reports with visuals. A significant feature is the ability to create and reuse 'skills' from successful workflows, eliminating the need for repeated prompting. Examples include an infographic correction skill and a YouTube description generation skill.
Nano Banana 2 offers a significant upgrade, especially for free-tier Gemini users, allowing up to 20 image generations per day. It boasts faster generation times (10-15 seconds) and improved aesthetics compared to the Pro version, particularly in consistent character generation and handling complex text in infographics. For complex prompts, 'thinking mode' is essential for accuracy, delivering perfect results where fast mode might struggle with minor errors.
Pomelli, from Google Labs, can create entire product marketing campaigns from a single image. Users upload an image, select templates, and the AI generates various photoshoot concepts. It includes editing features to refine images (e.g., removing a bottle lid) and then generates a campaign with taglines and visuals based on product DNA. The platform also extracts branding elements from websites for consistent campaign creation.
The Gemini sidebar in Chrome now integrates Nano Banana for image generation directly from web content, useful for quick edits or additions. A more impactful update is 'Auto Browse,' which allows Gemini to navigate websites, search for specific information (e.g., vacant land with criteria), and even interact with forms or shopping carts. While it can't open new tabs directly, it provides links to relevant pages.
Google Docs now offers an 'audio summary' feature, providing a quick voice summary of any document. The Flow platform, a creative studio for images and videos, has new features like drawing directly on images for changes and using 'ingredients' to generate specific video elements, like character entrances. Google Workspace Studio allows users to automate routine tasks across Google services (Gmail, Sheets, Drive) and external business apps using Gemini.
Opel, Google's Noode visual builder for AI workflows, received an upgrade with a new 'agent step.' This feature analyzes a user's goal, determines the best approach, and automatically calls the necessary AI tools (like VO for video or web search for research) to complete multi-step tasks efficiently. It aims to make node-based AI systems more powerful and user-friendly.
Google released Gemini 3 Deepthink, their most advanced reasoning model (Ultra plan only), competing with models like OpenAI's 03 Pro and Claude Opus 4.6 for complex reasoning tasks. Gemini 3.1 Pro is a slightly less advanced but much faster model, ideal for complex reasoning where speed is a factor. These models are designed for challenging tasks like building websites or interactive tools and dashboards, offering enhanced agentic planning and better results in Google AI Studio.