NotebookLM's image analysis unlocks new workflows for creative professionals

At a glance:

• NotebookLM's image sources, introduced in November 2025, now offer powerful multimodal understanding after the Gemini 3 upgrade.
• Users can upload JPG and PNG images up to 200MB each, with 50 sources per free notebook.
• The feature transforms how users manage design references, personal work, and random screenshots into a queryable system.

The overlooked feature that changed everything

For over a year, I've been a loyal NotebookLM user, relying on it for dense research papers, study prep, and long documents. I even tried bending it for tasks like managing bookmarks, thinking I had a solid grasp of its capabilities. The truth is, I had been ignoring image sources entirely. They've been available since November 2025 on the web and via a mobile camera button in early December, so they're not new. I simply dismissed them as a side feature for occasional charts or slide screenshots. Early image analysis wasn't impressive, and I never revisited the feature. Recently, however, I've built a workflow around image uploads and now use them more than any other source type.

This shift came from realizing that images aren't just attachments but primary sources. Instead of saving entire articles or videos, I now screenshot the exact reference material I need. This approach is cleaner and more efficient. Moreover, images don't 404 like articles and unlisted videos, making them a reliable archive. The upgrade to Gemini 3 in December 2025 was the turning point, as it brought significant improvements to multimodal understanding. Now, NotebookLM can actually see and understand the visual content in images, not just extract text.

How Gemini 3 revolutionized image analysis

Before the Gemini 3 upgrade, NotebookLM's image support was essentially OCR with a model layered on top. It worked well for text-heavy images like screenshots of articles or slides, but it struggled with visual content—such as UI screenshots without labels, diagrams, or photos of physical objects. The model would describe what it could read, not what it could see. The December 2025 move to Gemini 3 changed that. Now, the model understands images in both directions: it still performs OCR to index any visible text (making it queryable like a document), but it also comprehends the visual elements. For example, a screenshot of a Figma file is now treated as a real design source, not just an image attachment.

The technical details are worth noting. NotebookLM supports JPG and PNG images, with each source up to 200MB and a limit of 50 sources per notebook on the free tier. Google's help docs acknowledge that "certain types of images may not work as well," such as very abstract, low-contrast, or visually cluttered ones. In practice, I've encountered occasional issues where images are rejected without explanation, but these are rare. The key improvement is that Gemini 3 can now interpret the visual context, making images a first-class citizen in NotebookLM's ecosystem.

Building a workflow around image sources

My primary use for image sources is in a design references notebook. Instead of bookmarking articles or saving entire YouTube videos, I screenshot the exact UI patterns, diagrams, or breakdowns that matter. This is more efficient because a 20-minute video might have only 90 seconds of useful content. By screenshotting, I filter out the noise and ensure that NotebookLM treats only the relevant material as such. Additionally, screenshots don't disappear due to broken links, and OCR captures any visible text, so I can query for specific elements like "ghost button pattern" and retrieve the screenshot without remembering its file location.

The second part of my workflow involves screenshots of my own work. I capture Figma files at different stages, exports, and iterations that didn't make the final cut. This creates a queryable record of my design process. For instance, I can ask, "What did the onboarding flow look like before adding the progress indicator?" and Gemini 3 can answer based on the visual content of the screenshots. This has transformed how I track and reflect on my design decisions, making it easier to iterate and learn from past work.

Finally, I have a catch-all notebook for random screenshots—settings panels, error messages, or even physical objects like the back of a supplement bottle. This notebook has no theme, so I don't have to categorize images immediately. The mobile camera button feeds into this notebook, allowing me to capture anything from the physical world that might be useful later. What surprised me is how this notebook changed my behavior: I now capture more things because I know they'll be easily searchable and usable when needed.

Why it's worth catching up

Image sources have been available for months, and I'm late to the party, but the post-Gemini 3 version is now doing real work in my notebooks. Between design references, personal work, and the catch-all, screenshots and photos have become my most-used source type. If you've been treating images as a secondary feature, it's time to reconsider. The ability to query visual content opens up new possibilities for productivity, especially for creative and design professionals. While there are limitations—like the occasional rejection of images or challenges with abstract visuals—the benefits far outweigh the drawbacks. I only wish I had embraced this feature sooner; it has fundamentally changed how I work with information.