Nate's Blog

It looks like Google is about to re-add support for JPEG XL into Chromium, something they started working at a few years ago but quietly squashed. Ironically, I had just written about AVIF about a week before their announcement. It’s definitely a great development, while JPEG XL and AVIF are pretty comparable in terms of their ability to compress, JPEG XL has a few features that AVIF doesn’t.

Backwards compatibility is one reason you might choose JPEG XL over AVIF. You can convert a .jpg file to .jxl, and while the new .jxl file will be ~20% smaller, you can convert it back to the original .jpg at any point – shrinking your photo library a bit without actually destroying any data. It supports progressive loading, so it might be the best option if you want to share some higher resolution images on the web. It’s also capable of higher color detail and resolutions, although both formats can well exceed anything I ever expect to be using.

In regards to how well they can aggressively compress images, a while back I compared them for a potential blog post on file compression, taking notes on how they compared. Aggressively compressing images to the same size lead to slightly blurrier images in JPEG XL; AVIF being slightly sharper, but with slightly jankier lines (e.g. strait lines becoming jagged) and the colors seemed more likely to be desaturated or start banding. For now I’ll keep using AVIF for web images, if for no other reason than it’s the only one of the two supported on all major browsers, but it might be worth revisiting at some point once that changes.

Anyway, figured it was worth a share, either format can offer some pretty impressive compression on images. While compressing your best photos might not be the greatest idea – if you’re looking to store less important stuff (screenshots/informational pics), are looking to share photos online, or are building a platform and are looking for a way to keep the S3 bucket cost down – then either format might be worth looking into.

I'm experimenting with what I've deemed a 'mini blog' – smaller posts and different mediums separate from normal posts, while experimenting with different platforms/protocols, all linked together via a page and RSS link. If you want to see more you can follow here.

I haven't really used Google's NotebookLM since I toyed around with it last year, but I logged in the other day and realized they added a lot of new features to it – including a video generation option. I'm pretty late to the party, looks like it was added at the very end of July 2025, but it's new to me so I'm going to talk about it anyway.

The core of Notebook LM still remains the same: add sources in just about any form (documents, websites, YouTube videos, etc) and you can ask it questions or have it summarize the content. You can achieve similar results with a local LLM using RAG, but Notebook LM doesn't require technical know how to setup or a dedicated GPU. The only other feature I remember it having was a generic “podcast” generation, which while neat in theory, I didn't really see any uses for myself.

I might do a longer in depth post about it's various features sometime in the future, but for now I just wanted to share the video itself:

I generated the video with no prompt or guidance; I created a notebook, added this post as a source, and hit generate video. Honestly, I'm surprised at how well videos worked. It's no masterpiece, but it feels like it's escaped the orbit of AI slop and reached the level of generic human made YouTube slop. Though I'm sure being an AI summary of a human written text was a big part of it.

Anyway, I guess keep an eye out for that specific artificial voice on YouTube and similar platforms. I have a feeling that NotebookLM has/will be a major source of video slop.

I'm experimenting with what I've deemed a 'mini blog' – smaller posts and different mediums separate from normal posts, while experimenting with different platforms/protocols, all linked together via a page and RSS link. If you want to see more you can follow here.