An anime specialization VAE fine-tuned from sdxl vae with an anime dataset of 100k text-image pairs.
Description
FAQ
Comments (5)
Euge, sorry for asking, but what this VAE does to the images generated? I couldn't spot the difference on your example.
Also, thank you very much for making and sharing it with us!
The difference is slight. The line becomes softer maybe :)
Was this finetuned from the Stability VAE or the madebyollin fp16 fix VAE ?
fp16 fix vae
Cleaner results than normal sdxl vae or fp16 fix vae, but saturation is slightly increased, is this intended?