An attempt to create a realistic ANIMA model.
Version 2.0
Version 2.0 was primarily tuned for the "er_sde" sampler, steps 12-16.
Run: Euler/simple (Beta), CFG 1, Steps 8-16
Text encoder: https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/text_encoders
VAE: https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/vae
Description
A cleaner, more detailed, and highly realistic image.
FAQ
Comments (12)
Can you make this without the turbo baked in? I find the turbo lora is harmful to variety and quality.
Without Turbo, there's just regular 'ANIMA 1.0,' but the real value lies in the Turbo mode.
@toya_san so there's no changes to this model besides 'turbo'?
@CallMeMaybe There are several turbo lora mixed in a certain proportion, and a couple of models.
can you please leave up some prompt examples? i'm having trouble getting it photorealistic,
Photo of, realistic, real photo, photorealistic, masterpiece, best quality, 1girl, solo, long hair, black hair, blunt bangs, straight hair, round eyewear, cat girl, cat ears, cat tail, standing, hug, cat, looking at viewer, library
anima without any loras will make digital art that is a bit realistic if you just use realistic and photorealistic.
For actual real life photography you need to prompt for (real life photography:3)
That's just for base anima without any loras.
The model is simply phenomenal!
Would definitely recommend for anyone generating locally and transitioning from SDXL (IL/NAI) to Anima.
The original Anima sometimes produces broken poses when using too many tags, or too sensitive to artist style, while this model might seem a bit limited - it gives more stable results.
My favorite anima model so far! (and I'm generating anime art, not realistic)
Note for artist/anime use:
I found that using only one artist/style tag works the best, like `masterpiece, best quality, @fkey, 1girl, ...`
adding "anime illustration" to quality/style tags might override the artist tag.
It seems to be because of CFG: 1, but the speed boost it gives worth the limitation!
@Sluggish_Python Thanks for the tip. This is one of the best models ever.
Been testing ANIMA v1 with especially TURBO ones and with TURBO lora and this model is one of the easiest to use.
I'm so used to tag prompt and NLP is not my thing but this model seems to handle the tagging style just fine.
Also the Text render is really great.
Definitely using this one for now.
Thanks for the work!
Incredible. Thank you.