I am listing my requests and bug reports for NotebookLM. Simply writing them here won’t convey them to the developers or the community, so I am also posting them on the Discord community.
See also: Utilizing AI in Novel Writing—NotebookLM – Yashima Yugen
Table of contents
Feature wishlist (more to come)
Batch download
Download all sources and/or outputs in a zip file.
Dynamic sources
NotebookLM should be able to fetch latest information on a webpage or other cloud sources. It also fails to “reload” the newer content when a webpage is updated.
API
An ability to connect with other apps and services.
Creators Edition
The NotebookLM “Creators Edition” would provide more customizations for creators, and set a boundary between their creative domains and AI-generated domains. AI should not replace creators, but help them.
AI avatar (persona) creation
It would be nice to establish consistent characters to be used in outputs. An AI avatar, such as the one in Vids, can be a consistent character to be used in all visual/audio format. This process might be similar to character creation at the beginning of a video game. A company mascot character, such as the Bot (Bugroid), can be used for a business.
An avatar like this might be also accompanied by more consistent visual themes.
Voice Customization
At present, the Japanese output voice seems to only have two options: “Female” and “(Young) Male.” In English, there is also a male voice with a lower pitch. Being able to specify the speaking style and age would broaden the range of expression. If the podcast voices could also be customized, we could even have two characters from a story discuss the story itself. That would be fun. (• Discord | “Voice Customization” | NotebookLM)
Ruby (pronunciation guides)
Because Google Docs, which is used for source uploads, does not support furigana (ruby text) in Japanese text, one must take steps, such as converting Word documents containing furigana into a plain text format.
More on this topic: In the age of AI, ruby may be more important than you think – Yashima Yugen
Ruby characters provide the correct pronunciation for Japanese words. This feature is crucial for handling authentic Japanese texts, as ruby is commonly used in materials for language learners, children with limited knowledge of Chinese characters (Kanji), creative writing among others.
More about Ruby Characters:https://en.wikipedia.org/wiki/Ruby_character
Aozora Bunko is a volunteer-based website that hosts classic Japanese literature, and its texts frequently include ruby characters.
While the implementation of ruby support has been awaited in platforms like Google Docs for many years, it could also be implemented in NBLM. For example, it might be integrated by supporting the Aozora Bunko notation, which is one of the simplest methods.
Aozora Bunko Ruby Notation Example:https://www.wikiwand.com/ja/articles/%E9%9D%92%E7%A9%BA%E6%96%87%E5%BA%AB#%E3%83%AB%E3%83%93%E3%81%AE%E8%A1%A8%E8%A8%98
Supporting ruby characters would significantly help NBLM in two main areas:
- Output Quality: Rubies would ensure better audio and video output by providing the correct pronunciation.
- Translation Accuracy: They are also important for accurate translation from Japanese to other languages.
For the time being, I am required to upload two versions of the same text: one with rubies and one without. Implementing this feature would streamline the workflow. (• Discord | “Japanese text that includes ruby characters (pronunciation guides)” | NotebookLM)
For Chinese, pin yin pronunciation might be implemented in a similar way.
Output Cancellation
Currently, report and video generation cannot be canceled midway. Although this process usually only takes a few minutes, it would be great to have a cancel function. It would be even better if usage counts were not tallied upon canceling an accidentally started generation. (Please note: If you want to customize the output content, you must click the pencil icon in NotebookLM Studio. Be careful, as clicking the “Video Explanation” icon itself will immediately start the video explanation creation.) (• Discord | “Implement a Generation Cancel Button” | NotebookLM)
Fact-Checking the video script
The video explanations provide lively commentary, but occasionally, they “overstate” things out of enthusiasm, such as claiming something is “not at all present” when “a small amount is present.” It tends to use words like “absolutely” or “100%.” While the video explanation can be generated with just a click, it is time-consuming to correct the video if even one piece of information in the result is incorrect. We would like the script to be fact-checked before the output is generated. It would be also nice to have the video script in Markdown. (• Discord | “Fact-Checking: The video overviews exaggeration” | NotebookLM)
Mindmap in Markdown
Currently, the mindmap can only be output as an image. It would be good if the output could also be saved in Markdown format. (• Discord | “Mindmap as Markdown” | NotebookLM)
I have built an app as a workaround. More on this.
Report Output in Markdown
The report output is “formatted,” but for some reason, it is not in Markdown format, so I would like it to be converted to Markdown format.
Slide Deck/Video Customization
Slide Deck and Overview video would be better if it comes with an editable, complete pptx presentation, Google Slides, or Vids.
I have built another app as a workaround. More on this.
- Nano Banana specific issues
Nano Banana incorporates text directly into the image, but this behavior could be improved. When an image includes text, it is always a good idea to separate the text from the image. This is because a generated image with text often contains typos, garbled text (in Japanese and possibly in Chinese, even with Nano Banana Pro), and rasterizing fonts results in lower rendering quality. Furthermore, it would be even better if Nano Banana had an option to create editable image layers. OpenRaster or or PSD support?
Known Bugs
Shift_JIS text encoding
As of November 5, 2025, text files encoded in Shift_JIS become garbled when added as a source. These files must be converted to UTF-8 before uploading. (• Discord | “Japanese Shift JIS text is garbled” | NotebookLM)
Similar problems might exist for Chinese and other languages.