Type | |
Stats | 187 311 95 |
Reviews | (30) |
Published | Jan 31, 2025 |
Base Model | |
Training | Steps: 2,200 Epochs: 45 |
Trigger Words | hokuto no ken |
Hash | AutoV2 027B3524F0 |
Artist style lora by Hara Tetsuo, author of Fist of the North star.
Due to the dataset issue, if the character's distance is too far (ex. full body), there is a possibility that the face will be crushed, so we highly recommend using Facedetailer (or adetailer).
Recommend settings
V 1.0
MODEL : NOOB-Epsilon 1.1
step-30
cfg-5
dpm-2m / sgm-uniform
Use Facedetailer / Adetailer
V 1.5
MODEL : NOOB-Epsilon 1.1
step-30
cfg- 4 - 6
dpm-2m / sgm-uniform
Use Facedetailer / Adetailer
Recommended eyes prompt - intense eyes, *Constructed Puppils
*if you use Constructed Puppils, it's more badass-like.
Sketch prompt expresses a line more like the original-cartoon.
If the hairstyle or face, you want isn't working well,
you try adding "kenshiro" in the negative, it's likely to be resolved.
I recommend don't adding greyscale and monochrome prompts in negatives.
The character is not supported because it is a LORA whose purpose is to only train-style.
But, Kenshiro & yuria can be easily created.
UPDATE
- V 1.5
This time, Made it feel more like a cartoon if use a dpm-2m sampler.
If you don't like cartoonish lines, use eular_a sampler or use lora of Version 1.0.
I changed some of the datasets that I used in version 1.0, and added some female data. Now we can create a Yuria.
also create of char Lin, but it wasn't easy.
And the overall style is stronger than version 1.0, so use the weights between 0.6 and 0.8.