Type | |
Stats | 170 1,396 |
Reviews | (43) |
Published | Jul 10, 2024 |
Base Model | |
Training | Steps: 2,184 Epochs: 27 |
Usage Tips | Clip Skip: 2 |
Trigger Words | rainworldartstyle |
Hash | AutoV2 06EA7F6809 |
First, I want to thank Joar Jakobsson and Allegra 'Del' Northern the original artists for their amazing work, their efforts and the rest of the videocult team made "rainworld" the most unique game I played so far.
How to use: I trained the model on pony diffusion and it works with other pony models, I used clip skip 2 in training but it work sluggishly on none, trigger word is rainworldartstyle, the generated images in gallery were done through eurler a sgm uniform.
To generate like in game character selection art use "profile, black background".
I recommend 0.7 to 1 strength (try lower strength If it stucks to some sort of composition because of overbaking)
Use "slugact" if you want one, if you want none it is better to put slugcat in negative for a group use (multiple slugcats, group of number slugcat) this is still not perfect but can work.
Other tags that were included in captions include: (grass, flora, water, wet, wires, pipes, telephone sign, spear, rain, worms, waterfall, stalactites, blurry foreground, holding flag, monochrome, closed eyes etc).
This model can work well with img2img if text2img failed to produce desired results
Tip: putting pipe, human, girls and wires in negative can be useful if you dont want them and flowers look beautiful despite being not in captions, "lizard" "insect, bug" didnt work as intended...
This is work is my first lora experience, It really still needs refinement to tags, addition of the downpur dlc content and a lot of testing.
Neverthless it turned out better than I anticipated, civitai lora trainer is pretty good and combined with knowledge from articles like valstrix's guide it became much easier, the model can generate truly beautiful artworks.