v4 has been released! New tags have been trained "latex corruption, slime corruption, plant corruption, demonic corruption, tentacle corruption" give them a try!

Illustrious Version Is Here! Give it a Try

About

This is a LoRA based on the artist zxc. All images used were created by the artist. You can find out more information by visiting their discord here.

Using This LoRA

I recommend using this LoRA with a non-realistic style checkpoint model. The original art is of an anime/cartoon nature. Give the version notes below a read, as they will provide some information on use. This LoRA has transitioned from more of a concept of Latex Dungeon, to the art style of zxc's work. Prompting styles will be different between the versions.

Updates:

Illustrious v1: The same training dataset used in Pony v4 was used for the Illustrious version. They should act very similar in prompting. I recommend using between 0.7 and 1.0 strength. If you want as close to zxc's art style as possible, use at 1.0 strength and minimal prompt along with the checkpoint WAI-illustrious-SDXL. Take a look at the showcase images for prompt ideas and generation settings.

Pony v4: I have made quite a few changes for this version. The dataset has been pruned and optimized. The style is closer to zxc's art style than the other versions. This version also plays nicer with prompts and other models. Like v3, this is trained on not only the artist's style, but their game and character concepts as well. I have added more content from outside the Latex Dungeon game. I recommend using at 1.0 strength for this version. If you want less of the art style, try dropping it to 0.7 strength.

Pony v3: I have changed the prompting style of the training data. I have also changed the trigger word to "zxc_style". This version represents a change to embrace the art style of zxc, more than previous version. Previous versions focus more on the concept of Latex Dungeon

Pony v2: I have added a trigger word to this version and trained it in a different way to try and capture more of the style along with the concept. Seems to work well. Use it in the same way as the original version except for the addition of the key word. Weight recommended 0.7.

Pony V1.0: There is no trigger word. This version is based heavily off the concept of Latex Dungeon, and not so much zxc's art style. It tends to overwhelm other LoRA, and depends heavily on the checkpoint model. I recommend using V2.0 or V3.0.

tags to try: latex corruption, slime corruption, tentacle corruption, plant corruption, demonic corruption, laven, xenomorph, beemorph, tentacles, slime

i've found these settings work best:

Pony Versions:
weight: 1.0
clip skip: 2
samplers: "euler ancestral - normal" or "dpmpp_2m_sde - karras"
cfg: 6-7.5
steps: 24-30
resolution: 832x1216 (5:8 portrait)
negative prompt: due to most of the dataset being trained with grey and white backgrounds, i would recommend adding the following to negative prompt (grey background, white background, transparent background, simple background)


Illustrious Versions:
weight: 0.7-1.0
clip skip: 2
samplers: "euler ancestral - normal" or "dpmpp_2m_sde - karras"
cfg: 5-7
steps: 24-30
resolution: 832x1216 (5:8 portrait)
negative prompt: due to most of the dataset being trained with grey and white backgrounds, i would recommend adding the following to negative prompt (grey background, white background, transparent background, simple background)

My Other Latex Dungeon and ZXC LoRA:

Laven
Crown Demon
ZXC Slime Girls

Thanks!

If you like the LoRA give it a like and follow, and don't forget to check out zxc's discord linked above.