Added some examples with "stolen" prompts from other models.
Euler Ancestral only (all other look broken) o1.1 is different
This is a blocks merge among all official Anima models (from preview1 to Aesthetic 1.1). There are no training or LoRA usage.
For realism use this (long prompts break realism, sadly - the model needs training to support it properly):
Positive ("low quality" tag is not a mistake, but you can remove it):
(realism, realistic, photography), low quality,Also add "cosplay" word for an anime character.
Negative:
(anime, animation, disney), (@3dcg:0.8),
bad quality, simple background, mirrored, blur, AI generated,Anime style is default (I can do nothing with this without training).
Description
This one is slightly unstable. I may work on improvement in future.
FAQ
Comments (4)
It's incredible how good this checkpoint is. It has excellent prompt adherence, produces natural colors and poses, and feels incredibly consistent. I like it more than RDBT and WAI. It kind of reminds me of Illustrious, but with the benefits of Anima.
That's nice you found it useful. I'm experimenting all the time, so maybe something better will appear. It's not like training - it's mostly random. I never know how specific blocks from other models will affect a model.
Current model is a result of my pure curiosity about producing realistic images with Anima without training. Currently I have better prototypes with better backgrounds, but they have downsides. I want to find something good with minimal negative impact on other things.
That's a nice unique approach. I'd love to see what else you can create with what's available. I wouldn't mind seeing updates here and there.
massivly underrated checkpoint. gratulations. great work