The Breaker (Park Jin-hwan) Style

I really like the style of Park Jin-hwan, the artist of the manhwa 'The Breaker' and its sequels 'The Breaker New Waves' and 'The Breaker Eternal Force'. As well as 'Trinity Wonder', 'Promised Orchid', 'Merciless', and 'Absolute Regression'.

There are 2 versions of this LoRa:

• Version 1: It was trained with 515 images. They were screenshots of the webtoon 'Absolute Regression' and a few of 'The Breaker Eternal Force'. Due to the genre of the webtoons most of the images are of male characters and a few of female characters. The trained model was 'less-than-satisfactory/good-enough'. It was trained locally on my PC so the epoch and reps were very limited. It was tested with trained trigger word and other tags for best results. I mainly use Pony PDXL Negative Embedding (zPDXL2) and NegativeXL (D). The generated images using these 2 embeddings were average.

• Note: The embedding - NegativeXL (D) is for SDXL so it doesn't work for Pony models on-site and it would not be visible on A1111 Stable Diffusion locally to select it on the LoRa's list. A bit contradictory since I use PonyV6 as my checkpoint. So how can I use a SDXL embedding on A1111 SD? Simple, go to menu "Settings" > sub-menu "Extra Networks" > click on the option "Always show all networks on the Lora page" > click on the "Apply Settings" button (at the top of the page) > Go to your "LORA" sub-tab > click the "Refresh" button

• Version 2: Since I really didn't like the results of Version 1, I decided to look for a more specific training images mostly of female characters and I focused more on quality rather than quantity. This time I used 200 images for this version's training. 2/4's were from the webtoon 'Trinity Wonder' which has a wider variety of female characters. 1/4 were from the webtoon 'Promised Orchid' and 1/4 were from both 'Absolute regression' and 'The Breaker Eternal Force". The trained model was a lot better compared to Version 1. It was on-site which allowed me to have more epochs and reps. It was tested with trained trigger word and other tags for best results. I mainly use Pony PDXL Negative Embedding (zPDXL2) and NegativeXL (D). The generated images using these 2 embeddings were very good.

Version 1:

• Trigger Word: P4rkJ1nHw4n, manhwa, webtoon

• Weight: 1.0

Version 2:

• Trigger Word: P4rkJ1nHw4n, manhwa

• Weight: 1.0

Feel free to share your results on the gallery down below. If you can spare some buzz, I'd really appreciate it.