A. Azpiri Style

Last Update:2025-05-01 14:43:46
S
avatar_frame
Creations Number
0
Favorite
0
Download
0
Model Source:
Type:
LORA
Base Model:
SD 1.5
Trigger Words:
License Scope:
Creative License Scope
Online Image Generation
Merge
Allow Downloads
Commercial License Scope
Sale or Commercial Use of Generated Images
Resale of Models or Their Sale After Merging
Model Parameters:
Review:
0

Finally, the real second version of the LoRA is here. It has characteristics of the two previous versions (V1 and Soft). It's trained with Anything V3 and Kohya-ss utility, this time with 140 images (v1 was trained with 80 and Soft with 100).

The features of this version are:

  • With the LoRA at 0.9-1 of strength, the style is totally 100% Azpiri, similar to the printed comics, with very colored backgrounds, like on V1.

  • At 0.7-0.8 of strength, the style turns much more detailed and soft, just like Soft Version, but with less generic backgrounds. Some examples in the example pictures of this version.

  • Now it's as hot as Soft version, so you have to put the words 'nude', 'topless', 'nipples' and similar ones on the negative prompt if you want to do 'normal' sfw images.

  • As the Soft version, now the LoRA weights only 37 MB.

This time, like with Soft version, it's mandatory to use the negative embbedings EasyNegative and optionally badhandv4 or similar in the negative prompt to do good images. It's also recomended to write there (worst quality, low quality:1.4).

As in the previous versions, the best results are obtained using Anything V4.5 checkpoint. I recomend to use Hires.fix between 1'5 - 2x, with 10 hires steps, 0'3 or less for denoise and 'none' or Remacri Upscaler as upscaler. You can download Remacri upscaler from here.

translate
0 comment
0
0
0