Download
1 variant available
680 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 9
270 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 9
20 1 2 3 4 5 6 7 8 9
(8)
Feb 11, 2025
v1.0 - Illustrious: The larger training data set of the 1.0 version, trimmed slightly to reduce character repetition, and trained on Illustrious.
2860 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 9
1.1K0 1 2 3 4 5 6 7 8 9.0 1 2 3 4 5 6 7 8 9K
License:
Illustrious LicenseMy attempt to mimic the artstyle of Burbur, who was one of my favorite artists until she figuratively fell of the face of the Earth a few years ago.
If she request, I will remove this model.
v1.0 - Illustrious: Now for Illustrious! Also an improved training process (more epochs) and better model selection courtesy of local training. Still requires some additional negatives to suppress the watermark all the images had, see examples images for details.
v1.1 - Improved training by trimming the dataset to reduce character repetition and tagged characters & score ranges, hopefully this should improve performance and compatibility. Also expanded the concepts slightly by including some Pixiv only (i.e. not on Danbooru) images.
v1 - Recommended settings are a WIP, but initial tests show this LoRA works decently at a strength of 1.0, but may have compatibility issues with other resources. You don't need to specify "source_anime", but adding "source_pony" to the negatives seems to help. Also, using the quality tags "score_9, score_8_up, score_7_up," and the negative of "score_6, score_5, score_4," also helps improve the results. This is different from some other resources I've seen, which can work well without negative score tags.