Sign In

Kanae Kouda

49
372
2
0
Type
LoRA
Stats
287
Uploaded
Nov 21, 2023
Base Model
SD 1.5
Training
Steps: 10
Epochs: 10
Usage Tips
Clip Skip: 2
Strength: 1
Trigger Words
Kanae Kouda, medium length straight purple hair, blue eyes, yellow scrunchie with white polka dots, <lora:Kanae_Kouda:1>
Hash
AutoV2
5F6F0AC87D
0
0
0
0
0

Check out my Gumroad if you have any interest in buying specialty loras! https://satoulofthouse.gumroad.com/l/kvsoy

Created with 21 portrait images of Kanae Kouda with anylora checkpoint on Google Colab.

Version 2:

  • Still has trouble getting her scrunchie on the correct side but it's letting her have a full body and you can use different checkpoints for different situations now!

  • face doesn't get too much better from inapinting and look kinda watery. I chose save 1 for the version 2 update because save 10 (version 1) was having more trouble following directions.

  • Version 2 may actually have worse problems creating her hair after all. I recommend using version 1 to inpaint the face and hair and use version 2 to get something resembling her in the first place.

Notes about the data set, and what to expect:

  • You should be able to use Squeezer lora to make her have a body the suits her from the anime better, and to get it more likely to let you have more of a body and not just a head, try adding legs and feet to the prompt, however it will then, no matter the version, refuse to acknowledge that you are telling it to make her stand, and will insist on having her sit or have her knees up because it is trying to cram everything into a portrait shot.

  • If you tell it "yellow scrunchie" instead of the full name for it it will give the scrunchie, and if it is missing the polka dots add them back in with inpainting. If you do this, it will make it so it doesn't make every outfit she wears have polka dots.

  • Profile (side view) faces will look correct aside from the eye color being wrong. The eyes in the side view were wrong in the anime as well, and there are only two pictures in the data set with them.

  • I didn't think when I was tagging and forgot to describe the hair style, so there won't be much flexibilty in that regard. I will correct this later.

  • In the anime, the eyes were always just a bit skewed (as if the people drawing it didn't really know what they were doing) so if you wanna make it a little more even, you may wanna open an art program and use liquify to move one eye up a little bit. I'll doctor up the data set later to try fixing the initial pictures and see what I can do.

  • It has a heavy bias to try and make her wear bland clothes so create another character that has her body type first and then in paint the head and do some kit bashing to make her clothes and poses better.

  • There are no pictures of hands in the data set, so do whatever you need to while trying to fix those.