After testing local anime diffusion, I've decided to stick with NovelAI.

Why?

CivitAI has attractive models with impressive samples and a diverse style database, but there's a major flaw:
SDXL is fundamentally stupid model.

I believe the quality of the dataset and user efforts,through samplers, Lora workflows, etc. put into SDXL is significantly higher, yet they can't compensate for its shortcomings, which is disappointing given the dedication of CivitAI's developers.

Did you know that NovelAI comprehends prose exceptionally well? Even better than Flux!
I can input vague scene descriptions and generate an infinite variety of high quality outputs, all aligned with my prompts.

Pic related its a local image creation. It's quite beautiful, but it lacks a story or any significance behind its beauty.
If there's any beauty, it's in the quality of the checkpoints, which are often passed due to the diligence of the people from CivitAI.
And the AI artist prompts not with prose or a message with meaning, but aims to 'hack' the stupid language of SDXL.