V1

This is my first attempt at an Illustrious-based checkpoint, and I already see places where things can be improved.

If you know me, you know how much I tend to favor Hyper/LCM stuff, and indeed this checkpoint is a Hyper model. What that means is that you simply need to set the CFG to 1.5, sampler to either LCM or Euler A, and set the steps down to at least 8, and it should run fine. No tests have been done on other samplers, CFG configurations, or steps, so I cannot say how well it will work outside the prescribed settings.

Quality of the outputted images seems to improve when the following prompts are added:

Prompt: masterpiece, very aesthetic, newest, explicit, absurdres, nsfw

Negative Prompt: lowres, (bad), text, error, fewer, extra, missing, jpeg artifacts, low quality, watermark, unfinished, displeasing, early, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract], bad quality, bad anatomy, worst quality, very displeasing, oldest, safe

V2

That's right, Monet knows how to draw, Monet knows how to animate. Nevertheless, I used V2 to color in and fill the details on the above 2 hand-drawn lineart animations, using a controlnet.

V2 vastly improves on V1 in terms of quality, and like the previous version this is a Hyper-Illustrious model, meaning that one only needs to set the CFG to 1.5, and steps to 8, using either Euler A or LCM samplers to get good results. On the other hand though, setting the steps to as high as 45 doesn't seem to have negative effects. Realistically this model is for people like me who have not-so-great computers, where time is a premium. But, it seems to also work for people who don't read notes on my Hyper/LCM based models XD.