NotebookLM Now Generates Short Videos

Google has a long history of shipping products that feel genuinely useful right up until the moment they disappear. Google Reader. Google Glass. Google+. Stadia. The graveyard is well-tended and growing. So when a new NotebookLM feature lands with genuine capability behind it, the honest journalistic reflex is not to celebrate — it's to ask what Google actually intends to do with it.

NotebookLM's new "short video overviews" feature does something that would have sounded like marketing copy three years ago: you feed it your research notes, pick a topic, and it produces a roughly one-to-ninety-second video — the kind of thing you'd watch vertically on your phone, the format your teenager calls a "short" or a "reel" — complete with narration, animations, and motion graphics. No editing software. No script. One button.

The Futurepedia channel put the feature through its paces, and the results are worth understanding clearly, stripped of the enthusiasm that tends to accompany first-contact reviews.

What it actually does

The mechanic is simple enough to explain to anyone. You open NotebookLM, navigate to the Studio panel, and select "short" from the video overview options. The tool offers a few suggested topics drawn from whatever sources you've loaded into your notebook — articles, PDFs, research notes — or you can type your own direction. Then you wait somewhere between ten and twenty minutes, and a video arrives.

The Futurepedia presenter generated a video about earthquakes as a demonstration, and the narration that resulted was genuinely good — the kind of plain-English science explanation that a competent magazine editor would approve:

"Jagged rocky edges catch onto each other, completely locking the plates in place through sheer friction. Even though the edges are stuck, the rest of the plates keep moving. This forces the solid rock to warp, storing massive amounts of energy like a stretched rubber band."

That's not AI slop. That's coherent, accurate science communication. The animations reportedly matched the narration well enough that the presenter noted it "feels like someone actually made those motion graphics" — which, if you've spent any time watching the typical AI-generated video, you understand is not faint praise. Most AI video output still has the uncanny wrongness of a wax museum.

Other demonstrations covered everything from economic history (railway bubbles leading into the AI bubble — someone at Google has a sense of humor) to a book about an indigenous Mexican running tribe to a segment on Jevons paradox, the economic principle that cheaper computational power leads to more, not less, total consumption of it. The tool can be nudged toward narrative styles: the presenter asked it to reframe the fall of Rome as a true crime investigation, and the result held the structure reasonably well:

"The fall of the Roman Empire is history's ultimate unsolved murder, and the modern historical consensus of a peaceful transition — the dirt itself proves that's a massive cover up."

Whether that framing serves historical accuracy is a fair debate. The presenter's own observation on this point is the most honest thing in the review: Google appears to deliberately limit stylistic control in order to keep the output factually grounded. More control over presentation, the argument goes, risks more drift from the source material. That's a real tension, and it's not a stupid design choice — it's just a constrained one.

What it can't do

The limitations are specific and worth cataloguing, because they determine whether this tool is useful to you right now or useful to someone in a future version.

Style prompts don't work. The presenter asked for claymation. Got standard animation. Asked for anime. Didn't get anime. Asked for a whiteboard explainer. Not even close. Length prompts don't work either — a request for something under thirty seconds produced a video running over a minute. What you can actually steer is the topic and the narrative angle. That's it.

The output also isn't editable. When an animation produces a backwards foot or a plug that geometrically cannot fit its socket — minor artifacts, but real ones — you can't fix it. You generate again and hope. For personal use, that's tolerable. For anything you'd put your name on publicly, it's a problem.

Free accounts get three videos per day. Paid subscribers get twenty. The "cinematic" video format — a longer, more polished output — is only available to paying users.

The question Google isn't answering

Here's where I'm going to say what a product review won't: this feature is interesting precisely because NotebookLM is not Google's video platform. YouTube is. And YouTube Shorts — those phone-sized videos that now compete with TikTok for the attention of anyone under forty — is where Google has a direct financial stake in the short-video format.

The Futurepedia presenter noticed this too, and to their credit, said it plainly: "That also does make me wonder kind of where all this is headed since this is Google. I mean, are they testing things out on a journey towards a more generated shorts feed? One that could potentially even be like on-demand videos generated based on topics you're interested in."

I'd take that question more seriously than it's usually taken.

Google has watched what happens to advertising revenue when people stop searching and start asking AI. They've watched the same thing happen to video: recommendation algorithms that serve what you want before you know you want it. The logical endpoint of both trends is a feed that doesn't surface existing content at all — it generates content on demand, tuned to your specific interests and query. No creator. No copyright negotiation. No revenue share.

That future is speculative. But NotebookLM is a research tool with a surprisingly small user base relative to YouTube's scale, and Google is investing in making it generate phone-native video with real production quality. You don't do that because you want researchers to share earthquake explainers with their friends. You do that because you're building infrastructure and you want to see where the quality ceiling is.

What to make of it now

The feature, evaluated on its own terms, is genuinely useful in a narrow way. If you use NotebookLM to organize research — a subject you're studying, a book you're reading, a project you're tracking — the ability to generate a concise, well-narrated summary video is a real convenience. It's the difference between re-reading your notes and having your notes explain themselves to you. The earthquake example alone demonstrates that the output quality has cleared a bar that earlier AI video tools hadn't.

But "genuinely useful in a narrow way" is exactly where promising Google products tend to live before they either grow into something consequential or get quietly discontinued when a new priority surfaces. NotebookLM itself has survived longer than most, which suggests some internal commitment. Whether that commitment extends to developing this particular feature into something with real stylistic control and editing capability — or whether it's a signal-gathering exercise for a larger strategy — Google has not said.

What I'd watch: whether style controls arrive in the next two or three product updates. If Google gives users meaningful influence over how these videos look and sound, that suggests they're building a serious creative tool. If the feature sits where it is — useful but rigid — for the next year, that's a different signal entirely.

The tool works. Whether Google intends to let it keep working is the part they haven't answered.

Bob Reynolds is Senior Technology Correspondent at BuzzRAG.