Slightly softer lines and more neat details compared with original sdxl vae.
An anime specialization VAE fine-tuned from sdxl vae with an anime dataset of 100k text-image pairs.
