LoRA for Derpixon Artstyle.

This, in my opinion, adds a nice vibrant look to many pictures. Especially the multicolored gradients seen in much of Derpixon's work. It can be used for many things though keep in mind it may steer close to NSFW because of the nature of the source.

NB: consider putting (twintails, horns, animal ears :1.3) in negative prompts if you get strange stuff.

I trained it with 291 pics, which was a mix of mainly screencaps and some art. I trained it at 20 epochs with a dim/alpha of 32/16 with 1 repeat.

Stronger effect at higher LoRA weights, but can over power other LoRAs you have. Let me know how it goes for you. Read stuff below on guidelines and known issues :)

Guidelines:

• Start at LoRA weight 1.3 which is most consistent. If you are using other character/position LoRAs you may need to drop this style to 0.8 ~ 1.3.

• If you are not using other LoRAs then you can use this with a weight of 1.3 ~ 1.8.

• As mentioned previously, consider putting (twintails, horns, animal ears :1.3) in negatives. Especially at higher LoRA weights, as some bias exists with the training images.

• Works on different models though weights may need to change for good results.

• I used anyLora model and Midnight Maple model. Midnight Maple gave best results for this LoRA for me.

• Refer to my pictures for examples and what upscaler I used, keep in mind I have used other textual inversions to help with poses and negatives.

• I may or may not update this one day if I get around to removing some of the bias in the training images, but I think it works well enough for what it is to be honest.

• I'll add a list of the LoRAs I used over here sometime.

I don't think people read entire descriptions like this, but congrats if you do.