Sign In

Vladilena Milize

93
435
2
Verified:
SafeTensor
Type
LoRA
Stats
435
0
Reviews
Published
May 6, 2023
Base Model
Other
Training
Steps: 750
Epochs: 2
Trigger Words
Vladilena Milize
vmoutfit
Hash
AutoV2
E1A1B4D3F4
default creator card background decoration
DWeoss's Avatar
DWeoss

v1.0
This is my first attempt at making a lora model of a character. I accept any criticism, as long as it is adequate.

Of the known bugs at the moment - sometimes can generate the top of the main clothes of the character (at the neck), with what this is associated - unknown, appears quite rarely. Works with other lora models, I have not checked very much, but to work correctly, sometimes you need to lower the level of the lora model with which you want to use Vladilena, for example to 0.8, and then it's just a field for experimenting.

Tips: scale - 1 will be good

Training model: NAI

Used to generate models: AOM3, hassakuHentaiModel

Keywords: Vladilena Milize

Optional keywords: vmoutift (vmoutfit is a close to the original character's clothes, they are not 100 percent appropriate, but very similar. If you want to use other clothes, just ignore this tag and use something else, like hoodie, skirt, panties. If the clothes keep showing up sometimes even without the tag - try adding it to the negative prompt).

If you want better results, I recommend using 768 resolution. The model itself is trained at 768, on the models I generated the images I used not raw 768, instead I used a hires fix from 512 to 768 (in my case, on these models, hires worked better), here are the settings I used:

Upscaler: R-ESRGAN 4x+ Anime6B

Hires steps: 20

Denoising strength: 0.5 - 0.75

Upscale by: 1.5


If u know how to improve model, let me know, also report bugs that u find, i will try to fix them.