A VAE for generating higher dynamic and more natural image in comparison with the common VAE like 840000, Anything, kl-f8-anime2, etc.
Important note:
This VAE is for SD1.5 only!
This VAE may generate undesirable results on small resolution like 512x512, according to feedbacks. Color101 is more suitable for generating big image directly (that was what I did while tuning this VAE) or using hires fix for larger resolution.
The comparison images are compressed to .jpeg files automatically by Civitai.com, the difference of color shown here would be affected. So, it is better to make comparison by yourself.
Usage:
Put the file inside stable-diffusion-webui\models\VAE.
v1 update:
1. Provide more and clearer detail than most of the VAE on the market.
2. Fix detail distortion.
3. Skin tone is more natural than old version.
v0.1 update:
1. Fix green artifacts appearing in rare occasion.
2. Fix blurry detail.
What is VAE?
A VAE affects image where fine detail like texts and eyes present, at scale, it doesn't. If you look closely at the eyes in the comparison images, the eyes in some VAE sample has lost the slera in eyes. I recommend you that make some sample on your own to see the difference, because the images uploaded here were automatically compressed to .jpeg file with loss.
VAE in SD works as a decoder to calculate the latent space of a pixel for determining what color a pixel should be. The more complex around a pixel, the more factors needed to be decided, so in opposite, if your image structure is simple, it won't be changed drastically.
This Color101 VAE can give your images more vibrant colors, closer to HDR effect, higher saturation and add more redness to skin in comparison to other VAE on the market. If you like thick color, this should be your choice.
You can see https://stable-diffusion-art.com/how-to-use-vae/ for more detail.
Description
v0.1 update:
1. Fix green artifacts appearing in rare occasion.
2. Fix blurry detail.
FAQ
Comments (5)
can you give us more detaial about the vae model
A VAE affects image where fine detail like texts and eyes present, at scale, it doesn't. If you look closely at the eyes in the comparison images, the eyes in some VAE sample has lost the slera in eyes. I recommend you that make some sample on your own to see the difference, because the images uploaded here were automatically compressed to .jpeg file with loss.
VAE in SD works as a decoder to calculate the latent space of a pixel for determining what color a pixel should be. The more complex around a pixel, the more factors needed to be decided, so in opposite, if your image structure is simple, it won't be changed drastically.
This Color101 VAE can give your images more vibrant colors, closer to HDR effect, higher saturation and add more redness to skin in comparison to other VAE on the market. If you like thick color, this should be your choice.
You cam see https://stable-diffusion-art.com/how-to-use-vae/ for more detail.
eyes and faces are getting all messed up when using this VAE on my checkpoints. Shame, the colors are really good.