I am sharing with you a 3D model that is very close to realism. Does not require VAE. "Clip skip" gives successful results in 1 and 2. Depending on the prompt you will use, it may approach realism, 2D, 3D or anime styles. When you use it with Lora, set the "cgf scale" value to 5 or 6. For better faces you can use "After Detailer". It gives the best results with "Euler A". I currently have 3 different versions of this model. If you like, I can share them too. I would be happy if you share your ideas and comments with me.
Note: The resolutions I use in the images I share are mostly 512x768 and 640x960. I usually make them 1024x1536 using Hires fix.
Note2: Tends to create images of pleasant and attractive women
Description
Vae included, pruned version.
FAQ
Comments (8)
I like this alot, well done.
Thank you :)
Can you please explain how do you get such crispy and iper detailed images? Never seen such quality before!
...And is just SD 1.5? I'm shocked! O_o
Thank you for liking it. I combined my BeautyFool model with some photorealistic models that I trained and did not publish. The result was really great. I used "hiresfix" between x1.6 and x2 on all images, and used "after detailer" to enhance faces in non-close-up images.
@53rt5355iz no problem, I always use 2x and Adetailer. Now testing...
@53rt5355iz Tested on 120 images with various subjects and I prefet not to vote it in respect of your work. But I think needs some tuning. The super contrasted colors have pro and cons: sometime give nice images, sometime burns details. In general this "electric" contrast looks cool on portrits but give poor definited faces at full body images. Another problem is there are often wrong body proportions, expecialy if you try to set a different age for the subject. Success rate is just medium, you get good results but after lots of images to drop. At the moment, my opinion is it doesnt reach the quality and succes rate of your BeautyFool.
@53rt5355iz To be honest, I noticed is very dependent by your complex negative prompt. Things worked a little better when I copied it, but I have only verybadimagenegative 1.3 installed, so I dont know if having the othere two would make a difference
@settima_ai Thank you very much for your time and comment.