V3
A completely new model that is very different from V1 and V2. The dataset was created entirely by me, and the images in it are in very high resolution. Many popular characters were used in the creation, so the faces and body anatomy had to be as close as possible to what we see in anime.
This version has much more contrast, subtle contours, and is perfect for both SFW and NSFW images.
The goal was to create lora that would blend well with other loras and complement the style tags perfectly. This is also great if you simply want to make a beautiful anime screencap image, as without loras at standard checkpoints, anime screencaps often look either hyper-realistic or lacking in detail.
POSITIVE PROMPT: anime screencap, anime coloring
NEGATIVE PROMPT: 3D, realistic
V1 - V2
Created using anime images from the internet. I happily scavenged and edited them, but the low resolution sometimes made the generated images look blurry. Dataset: 200+ images
Description
new high-res dataset
Better characters anatomy and faces
more vibrant colors
Better NSFW
Subtle contours
Text bubbles works better
Much better prompt understanding while creating 2+ characters on one pic
FAQ
Comments (3)
There's a typo in the filename: instead of "Better" it's "Batter" (I only noticed because I was doing XYZ plots and it wouldn't load it...)
Oh, sorry, I probably wrote the wrong letter in the name(
@16_5 No problem, I mainly noticed because I was changing "V2 to "v3" when testing your latest LoRA (also my bad for not checking the filename before testing...)