Simple conversion of Crody's Nova Anime AM v4.0 model. All credit to them.

Still my favorite Anima base. works incredibily well without the need of quality tags while still providing a good basis for style tags and LoRAs.

Why int8 when the model is already so small? Speed.

Int8-convrot provides around a 50% speed-up in generation times on my RTX 3070 (39s vs 56s on BF16) with virtually no decrease in quality.

tool used for conversion: https://github.com/Starnodes2024/comfyui-starnodes-modelconverter

my recommended settings for fast generation:

ER SDE - simple - cfg 1 - 16 steps - Turbo LoRA @ 0.5 strength