General Info: This is a remake of my SD1.5 LoRA in Pony with an entirely new dataset. Each of the 75 training images has been handpicked, manually cleaned and cropped. Trained on base Pony for wider compatibility. This LoRA is based on the works of _dabdabil (aka DabiBill).
Activation word: none. Strength: 1. This LoRA works good both for SFW and NSFW. ADetailer is strongly recommended (at least for the faces).
Style Description: cartoonish digital 2D inspired by traditional art (varied brushes, mixed styles). Characters (especially if prompted as young) have thick limbs and fingers. Expressions are both diverse and easy to prompt.
Weaknesses: overtrained character LoRAs can impact cohesion (use BREAK command or drop their value). Faces and male "parts" in the distance tend to get smeared if you don't use ADetailer.
Description
FAQ
Comments (4)
Updated
More like "remade", entirely new dataset. I was trying eliminate artifacting that was present in the previous LoRA, so I used generative fill in Photoshop to remove everything in the images that is not the character. You know how his works resemble collages. Overall, it worked, you only sometimes get this small specks around the character, but I think it's just due to the original brush style.
@Noise_Connoisseur u made a well made lora, other than it doesnt look as much as his style, its really good and looks different like a unique style lora
@xmattar thank you. It was a personal stylistic preference, roughly 40% of the dataset are drawings from late 2020 and early 2021. They aren't as "chibbi" as his later work, hence the look.