Long time no see?

Not much to say compared to how long I've been quiet.

As everyone knows, the new trend has shifted to [anima], and I've been using it happily.

Also, though not recently, I upgraded my graphics card, so my work environment has improved, but since real life has been busy, I haven't been able to put much time into model development (and it'll probably stay that way for a while).

The base version of [anima], or even 1.1, is plenty satisfying, but I guess it's just habit or maybe my hipster tendencies I couldn't just sit still.

I've upgraded my GPU and such a great model as [anima] has come out, I really wanted to tune something myself. But gathering a dataset takes more time than expected, and I'm pretty lazy. :(

So, once again, I couldn't resist and ended up merged one anyway...

My approach hasn't changed much since the old XL days. Starting from the base, I make it a bit more stable and just slightly better. So I want to mention upfront that this model's knowledge level doesn't stray too far from the base [anima] model.

Usage

Honestly, for the most part, you can just follow the same settings and prompts used for the basic [anima] model. No need to go into a long explanation here.

That said, here are a few recommended settings:

Sampler: Euler A CFG++
Scheduler: AYS
CFG: 2–3.5
Steps: 16–28

Thanks to circlestone_labs (tdrussell) for creating such a wonderful model.

Anitional (Anima Checkpoint) - v1.0-int8c