Type | |
Stats | 2,476 16,214 |
Reviews | (346) |
Published | Feb 10, 2024 |
Base Model | |
Training | Steps: 3,400 Epochs: 5 |
Usage Tips | Clip Skip: 2 |
Hash | AutoV2 4448ABD1EA |
This is a lora trained with about 130 images from the Disgaea series. (Disgaea 5/6/7/RPG) My goal was to capture the flat, 2d style the games are known for. It's not a perfect re-creation due to the influence of the model and the variety of images I used, but I am happy with the results nontheless. Please note, this is not a character lora. My goal wasn't to recreate the characters from the game, it was only to capture the artstyle.
This lora was trained on Autism Pony. I haven't tested it with anything else.
To get the results you see in the images above, these are the settings I used.
Prompt: score_9, [your prompt] <lora:DisgaeaXL:0.8> (using '1' feels too strong)
Negatives: mosaic censoring,bar censor,censored,source_cartoon,source_pony,source_furry,
Sample: Euler a
Steps: 33 (Whatever gives you the best results)
CFG Scale: 7
Highres Fix: 4x-UltraSharp
In most of the examples I intentionally used tags like pointy ears, and devil eyes to give the images that "Disgaea" feel. They are not features inherit to the lora itself, but they were a part of the training data. All of the images I used for training were sfw, so the nsfw images aren't as strong as I'd like.