NotebookLM has been the reigning champion of research-oriented AI for three years running, without a worthy competitor in sight, and it just keeps getting better. In the last year, I've spent a considerable amount of time curating notebooks on the platform, one for every topic I need to understand better, save for posterity, or make searchable and podcast-able without suffering the ills of hallucinated results.

However, power users have moved on from generating customizable podcasts and interrupting them for questions like a live listener. Since January this year, Google has announced and added three important features that make NotebookLM way more customizable than before, a treat for visual learners, and the real Swiss Army knife you might need for the next assignment the uni throws your way. Cinematic Video Overviews, customization for mind maps, and the ability to generate desired file formats from multimodal sources are here to change the way we use NotebookLM.

Visual learning on steroids

A documentary approach to your personal notes

When Google first introduced Video Overviews earlier this year, I honestly expected a revolution in visual learning, but we mostly got a glorified version of the Slides feature in the Studio sidebar, narrated in a robotic monotone. I remember trying to synthesize my camera's dense exposure bracketing instructions, only to receive a flat, unimaginative presentation that suspiciously resembled a corporate PowerPoint from 2004.

The new Cinematic Video mode is in a different league, though. It turns your uploaded research into fully animated, documentary-style explainer videos. Powered by a heavy-hitting technology stack of Gemini 3, Nano Banana Pro handling the visuals, and Veo 3 driving the video engine, it makes better decisions about pacing, visual style, and narrative flow of the generated video. Using this to decode my camera's custom button assignments was perfect for visual learners like me. Add to that the text box, where you can specify topics of interest, your preferred art style, and any other details you want covered.

However, this shiny new toy comes with an incredibly tight limit on everyday utility. You can only generate two such videos, even on the Google AI Pro plan. Unlimited access will cost you the steep $250-per-month Google AI Ultra subscription. Moreover, my colleague Mahnoor found that if an AI hallucination causes a generation to fail midway, you simply lose that slot, making it painfully hard to iterate on prompts without burning through your precious daily allowance.

Steering the visual chaos

Building a custom map of your ideas

Mind Maps inside NotebookLM used to feel like a strict take-it-or-leave-it feature that lacked any real user agency. You upload a mountain of sources, select the ones needed for the Mind Map, and hit Generate. The AI typically figures out its own web of connections among important points and topics mentioned in the sources, but I've seldom found the sequence of nodes to be accurate.

Sure, Google AI gets the nesting right with related concepts grouped under a relevant parent node, but these parent nodes aren't sequenced correctly. This often resulted in broad, generic overviews that were completely useless when I needed to drill down into a specific technical issue in the source documents. Now, you can steer the system with precise text prompts, forcing the mind map to focus on the exact angle or problem you need solving. Instead of getting a top-down view of the entire camera manual, you can ask for a map of just the features you are most likely to use.

I immediately tested this by asking NotebookLM to map out purely the low-light limitations of my Tamron lens across my different uploaded reference materials. The system categorized the nodes into a clean decision tree, highlighting high ISO noise thresholds on one branch and autofocus failure points on another, saving me hours of manual reading. Yet, for all this newly acquired intelligence, the execution still lacks perfection. If the AI gets the sequence of nodes wrong, you cannot click and drag to reposition them manually. You are permanently stuck with the rigid AI-generated layout.

The ultimate file factory

Turning notes into finished deliverables instantly

Perhaps the most game-changing feature quietly creeping into NotebookLM is its upcoming ability to generate your desired file formats directly from supplied multimodal sources. Google's previews suggest we're moving past the tedious era of querying a chatbot, copying the result, pulling up a new document, and wasting another 20 minutes formatting it to fit the desired layout.

With Gemini 3.5 Flash's processing horsepower, we can command Gemini to turn a messy folder of sources into a clean spreadsheet, a polished briefing document, or a presentation deck. Conversion will be a native feature and should run four times faster than previous versions. I'd also recommend asking the AI which file formats are best suited to representing information from the selected sources if you're unsure which format to pick.

This upcoming capability could streamline complex workflows, especially when you consider how adding Studio panel outputs can be used as sources. Tying this to the customization options already available in Mind Maps, you could imagine how flowcharts could easily enhance slide decks and formal reports. Because it strictly pulls only from the grounded materials you provide, there is a near-zero risk of it hallucinating fake data or talking points.

Replacing several tools in one shot

Collectively, tying all these powerful tools together reveals a platform that has finally matured way past its experimental, viral launch phase. The Cinematic Video Overviews provide the rich, animated visual context that dense technical manuals have desperately lacked for decades. I might still harbor a healthy dose of skepticism when massive tech giants over-promise on their roadmaps, but this current iteration of NotebookLM is genuinely improved from a functional standpoint.