Lora trained on the art style of Hanz Kovacq.
WARNING: the dataset contains NSFW! ADAPT YOUR PROMPTS TO AVOID NSFW CONTENT!
This lora is intended to capture the art style of artist Hanz Kovacq. It can reproduce the style of his black and white comic panels and of his colour drawings. The lora can also generate some medieval stuff (as featured in some of his comics).
Possible issues to consider:
The dataset contains many NSFW images. This artist draws erotic comics and I had little material to work with. Using mostly NSFW images was inevitable. It is still very much possible to draw good SFW stuff, but you definitely need to adjust your prompts (see my examples).
The faces of b&w images are of lower quality than the colour images. This can be solved by using Adetailer, img2img, upscaler, etc...
The lora is not as flexible when it generates colour images. B&w images also tend to have better compositions.
The lora tends to generate women, despite an evenly mixed dataset. This can be fixed with the right prompt (see my examples).
This lora gives better results if you use it in combination with a more generic style lora. See my examples. (This helps it overcome the limitations of the dataset). Results remain consistent enough in style.
This lora is not trained on full comic pages.
I was able to make some good knights in armour, but things fall apart when you try to add a sword. Some medieval clothing and settings are also possible.
I was not able to recreate the 18th century clothing that I tagged in the dataset.
I avoided all mutations/bodyhorror (this is present in some of his comics). There was still some bdsm in the dataset but it was all tagged. So best to add "bdsm" to the negative prompt.