The result of many merges aimed toward making extremely highly detailed images. This can do realistic content, but I prefer to make stylized stuff so that aspect has not been extensively tested.
All example images were generated without any hypernetworks, loras, or inpainting. Everything was genned without high res fix, then upscaled using SD upscale in img2img.
Used SD 1.5 VAE
Unpruned float 32 version of AyoniMix V2 on huggingface:
https://huggingface.co/Ayoni/AyoniMix_V2/blob/main/AyoniMix%20V2%20Float32.safetensors
Description
Now with VAE baked in.
FAQ
Comments (6)
Hello! Thanks for the great upload. I used V4, but no matter what VAE I use, the images are monotone, very washed out. I have like 10 different VAE. And the prompts work as usual with other models. The images generated look really good. it's a pity to delete it. Please let me if you have any thought about that happened. Thanks!!!
I went ahead and uploaded a version with a VAE baked in. The VAE I use is the standard SD 1.5 MSE VAE. I've baked that one in, and I also uploaded it seperately as an attachment to the version without a VAE baked into it.
Thank you very much!!! Works right out of the box! You are a genius!
v3 was 7gb and v4 is 2gb
does that mean v3 version is better?
The opposite actually, it's better for it to be smaller for a number of reasons. All SD based models have the same number of parameters at 859.52 million no matter how much mixing, pruning, or dreambooth training you do. When you prune, you are taking out the extra precision for FP32 which saves 2gb. You can also remove the training weights which is useless unless you do large-scale training (99.99% of people are not going to do this as it is very expensive and time consuming) which saves another 2-3gb.
Trimming all that extra stuff out saves VRAM, storage space, and makes the model load faster without any loss of quality in the generations. I've moved to only using pruned models as it allows me to mix them much faster and not use as much space.
oh, that's way better then!
I thought the bigger the filesize the better cuz it has more information.