I've been a loyal NotebookLM user for over a year now and have never had a real reason to look elsewhere. Like everyone else, it's how I get through dense research papers, prep for studying, and work with long documents I don't have the time or patience to read straight through. I've also tried bending it into shapes it wasn't really designed for, such as managing bookmarks. So I thought I had a good handle on what it actually does well.

What I'd been ignoring this whole time was image sources. They've been around since November 2025 - so not new or a hack per se. I just had this default mindset that "real" sources are documents and links, and images were a side feature for the occasional chart or screenshot of a slide. Image analysis also wasn't amazing when it first rolled out, from what I remember, so I wrote it off early and never really dove back into it.

Lately, though, I've been leaning on image uploads way more than any other source type, and built a proper workflow around it.

I'd been sleeping on the most underrated source type in NotebookLM

Image uploads aren't just a side feature

Image sources hit NotebookLM on the web in November 2025, with the mobile camera button following in early December - a floating action button that opens straight into a new source. So this isn't a recent feature drop, and I'm definitely late to the party. In December 2025, Google moved NotebookLM from Gemini 2.5 Flash to Gemini 3, and the announcement specifically called out "significant improvements to reasoning and multimodal understanding". The multimodal part here is where the model actually reads images rather than just text inside images.

Before that, image support in NotebookLM was basically OCR with a model layered on top of it. Fine for screenshots of articles or slides, or anything heavy on text. Less useful when the image itself was the thing, like a UI screenshot with no visible labels, a diagram, or a photo of a physical object. The model would describe what it could read, not really what it could see. Then with Gemini 3 in the backend, image sources now work in both directions - OCR still catches and indexes any text in the image so it's queryable like a document, but the model also understands the visual content. So, for example, a screenshot of a Figma file is a real design source now, not a glorified image attachment.

A few quick specs worth flagging: JPG and PNG are supported, up to 200MB per source, and you get 50 sources per notebook on the free tier. Google's own help docs do note that "certain types of images may not work as well", which is worth mentioning - very abstract, low-contrast, or visually cluttered images can still trip it up. And sometimes it just doesn't accept my images as sources for no apparent reason.

My design notebook is basically a queryable Pinterest now

Where my references and my own work both live

I've been collecting design references the way most people do for years, with bookmarks, an old Pinterest board, and screenshots saved into folders I never revisit. And even if I did actually revisit these references, none of it ever talked back to me. So now when I'm reading an article on a UI pattern or watching a YouTube breakdown, I just screenshot the moment that matters and add it to my design references notebook instead of adding the full article or video.

Screenshots beat full sources here because they're already filtered. A 20-minute YouTube video has maybe 90 seconds of actual reference material, but adding the whole thing means NotebookLM treats all of it as equally relevant when I query. Same with articles. And screenshots don't 404, which articles and unlisted videos very much do often. OCR catches any visible text too, so I can ask "where did I save that ghost button pattern" and get it back without remembering which file it was in because NotebookLM can see the tab name of the project I screenshotted.

The other half of this notebook is screenshots of my own work as it evolves - Figma files at different stages, exports, and iterations that didn't make the cut. It gives me a queryable record of my own design process. I can ask "what did the onboarding flow look like before I added the progress indicator", and because Gemini 3 actually sees what's in the screenshots, the model can answer.

The "I'll come back to this" notebook

Where random images and screenshots go to be useful later

My screenshots folder on PC syncs to Google Drive automatically and it turned out to be the thing that made this workflow viable. So when I screenshot something I want to deal with later, like a settings panel or a specific error message, it's already in the cloud and I can fetch it from within NotebookLM whenever. The trick is having a notebook deliberately called something like "figure this out". Most of my notebooks are topic-specific, but this one isn't. Stuff gets added, queried, sometimes migrates to a proper topic notebook later, sometimes just sits there until I need it.

What makes this notebook actually useful is that it has no theme, so I don't have to decide what something is the moment I capture it. The mobile camera button feeds into the same notebook too so anything in the physical world I might want to come back to goes in (like the back of a supplement bottle or a confusing label on something I just bought). And what I didn't expect was how having this notebook changed what I bother capturing in the first place.

I regret not doing this sooner

Image sources have been around for months now and I'm pretty late to utilizing them properly, but the post-Gemini 3 version of this feature is doing real work in my notebooks now. Between design references, the in-progress work, and the catch-all notebook for everything else, screenshots and photos have become my most-used source type. If you've been treating images as a side feature in NotebookLM the way I was, it's probably worth catching up.