Parable of the Rifle and the Machine Gun The drawing AI to devise and generate each prompt one by one by a human is like a rifle that aims and fires one shot at a time. What I created in C3Cats is a system that automatically fires a series of shots after aiming. It is like a machine gun. It is likely that the 40,000 rounds shot so far have developed an implicit understanding of “how bullets move. The equivalent is an understanding of the behavior of the drawing AI. NovelAI can’t use the same continuous firing feature, and unlike C3cats, it often tries to “get what it’s aiming for”. I realized that this is a different application, but I sometimes unconsciously and naturally make a judgment that “from here on, it is better to simply fire in rapid succession instead of devising a prompt”, and where this judgment criterion came from was my first-hand experience of bullet movement obtained through machine gun firing.

Major update yesterday I was able to solve the problem of not being able to update the Thompson sampling data, so I updated it, but I get warnings about missing data files in several cases. I can ignore it and continue, but what is this, did I break something yesterday? →Mixed images generated by NovelAI, which does not have a StableDiffusion-derived data file in it.

That’s true. I don’t see any real harm in leaving it alone, but it’s weird to get a warning


https://twitter.com/nishio/status/1580559634745917440?s=46&t=LPYVim0EsaY4jqUJjzD0-Q

https://twitter.com/suyakdn/status/1580890937953116161?s=46&t=jO4GIL-1mQgimJ5TeQE5kg

https://twitter.com/wander00317/status/1579796209593888768?s=46&t=jO4GIL-1mQgimJ5TeQE5kg

image


This page is auto-translated from /nishio/旄蚘2022-10-16 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.