Hi all!
It's been a while since I've had an update here. Long story short: after our previous host repeatedly shut down the site at the beginning of the year, I stopped everything and worked on switching to a new host. The Big Switch happened back in February but since then I've been working behind the scenes to move the rest of the servers and services over, modernize our code and system architectures, and make sure the site would be stable on our new host for years to come.
However, as of a few days ago, all of that migration work is finally done!
And now that I have more time to work on features and bugs again, I wanted to start with a big one people have complained about: when we moved to the new host, we had to disable Basil's image visualization for your notebook pages. Long story short again: the model we used for image generation was too old to work on the new systems, and I didn't have the time to hunt down for another equally-ethical model to switch to. However, as I just said I have a little more time now… so I'm happy to say I'm re-enabling image generation on the site with a new model version which is focused on photorealism across diverse character demographics, excludes copyrighted data from the base LAION dataset many other image generation models use, and has been fine-tuned with public domain and Creative Commons images to further distance its latent space away from copyrighted works and focus more on “first draft” and conceptual visualization, rather than “finished art”.
Two things I want to reiterate with this model (which are also true of the previous model):
- No data from Notebook.ai was used in any way (training, fine-tuning, etc) in the making of this AI model
- Since we're purposefully avoiding the big copyright-laden models, our quality may not always be as good
I've done my best to test this new model across an extremely diverse set of my own characters and notebook pages, but I rely on the image rating system in Basil to ensure that image generations depict characters (and other pages) of all backgrounds equally well – so if you try out this new model please do rate the images generated so I can get feedback on what improvements still need to be made.
To celebrate the return of Basil, and simultaneously apologize for the image generation feature being disabled for so long, and also help collect more ratings from the get-go, I'm planning to open up Basil's image generation to everyone, unlimited, for free, for the entire month of June. During this time, you'll be able to generate an unlimited number of images for all of your pages on Notebook.ai. I'll post another announcement about this with more details soon.
In the meantime, here are some additional links that might be handy:
- Basil's page on Notebook.ai, where you can start generating images from
- Our stats page for Basil, now filtered to just show this newest model version (old version here)
- A long-form version of this announcement, over on our blog, with some example images
- Our original blog post about how this feature works
- Notebook.ai's Commitment to Ethical AI
But yeah, I'm back and I'll be working through your reported bugs and suggested features again very soon – and I'm excited to be back. :)
Happy worldbuilding!