Denizens of this wastebin, I bring you culmination of my suffering, frustration and anger.

After training a network on the peak of 1990's trash such as Venus Wars, Jin-Roh, Slayers, Yu Yu Hakusho, Evangelion, Stardust Memory, Berserk (The 90's one), Akira, Ghost in the Shell and Cybercity Odeo 808, I have produced a style lora, after much pain.

This piece of jumbled uncaptioned mess, has a great tendency when accompanied by the tag "1990s \(style\)" along with its trigger tag "tr90style" to produce the peakest of weebtrash from the golden age of Anime.

This unsorted mess produces oftentimes great results at weights close to 1.0, be it 1.2 or 0.8, your mileage may vary heavily depending on the base model, the CFG, clip skip (use 2 you Neanderthals) and the constelation of the planets in regards to mercury. Works best with closer in shots, as that's what the dataset was biased for, though I am sure you can get it to work.

The usual suspects for suggested checkpoints are anything Meina, StoreBoughtGyoza (That one in particular really gets the aesthetic), and any anime model with thin lines and generally training done on shows, or not heavily stylized bases. (Bunnyspoon injects too much of its own fluff in)

Have fun, share generations and someone please for the love of god write a good tutorial on how to make style loras/lycoris specifically.

I will upload the 4th and 3rd epoch LORAs, had good results with both and you should be intelligent enough to pick yourself.