Name3000 style ILLUST

The style of Note3000.

Trigger word is name3000. The source material are actually AI-generated images, making this a second-order LoRA. The training tags were entirely autogenerated due to the unusual dataset source, so the concept knowledge may be slightly worse than in my other LoRAs. I compensated for that by picking an earlier epoch than usual. If you encounter undesirable themes in your generations, I strongly suggest using corresponding tags in the negative prompt.

I stumbled upon this style on one of the lesser-known booru-s, and some research led me to the origin of it -- a series of 4chan posts. The author actually described their technique (NovelAI v4, s16xue, zankuro, 3d, [[fizrotart]], [[[khyle., tom bagshaw]]], [[[[ciloranko]]]], 2023), but when I tried similar combinations in Illustrious, it didn't feel the same. So I hand-picked and downloaded some samples from thebarchive, auto-tagged them and trained a LoRA. Since the style is already pretty close to the model's area of knowledge, it took less epochs than usual to achieve satisfactory results.

If you like the style... I don't know, go on 4chan and say thanks? There's no way for me to link to the original author that I know of (are there even user profiles on 4chan?), and I absolutely do NOT encourage supporting NovelAI (paid services aren't bad per se, but I don't believe they deserve community support when free alternatives exist).