Consider adding danbooru/e621 artist names to the prompt words in the training set to obtain more diverse styles？
I tried this model and it's great, but as the title says, it's too one-dimensional and prompts for artist names or specific styles don't work for it（when generating anime images）. Adding danbooru/e621 artist names can create more diverse content or even create entirely new styles, like NovelAI's anime model/Illustrious/NoobAI
bit late for that now fam, can't just add random shit to the dataset mid training
it was likely never trained on any specific artists
I agree
@ruleez
.
If people want some specific styles they will surely be able to create LoRAs or fine-tunes when it gets stabilized. I myself is nowhere interested in anime style.
I trust the creator(s), this model has demonstrated impressive progress already, let's give it some time.
And I wait for each new version like a kid in a candy store...
The dataset can and has already been adjusted multiple times throughout training, but it's not needed for artist tags:
https://www.reddit.com/r/StableDiffusion/comments/1j4biel/comment/mg7x4yp/