I decided to make an Illustrious model containing all five members of the mermaid band Drowned Sorrow, a series of OCs created by me and <a target="_blank" rel="ugc" href="https://civarchive.com/user/Pizzamaster98">https://civarchive.com/user/Pizzamaster98</a>. On rhythm guitar and main vocals, Thalassia Van Darkholme. On lead guitar, Marisa Milos. On bass and backing vocals, Dana Herrington. On drums, Julienne Wolf. On keyboard, Venus Lee.This is a successor lora to a series of previous models I made:<a target="_blank" rel="ugc" href="https://civarchive.com/models/774290?modelVersionId=866020">https://civarchive.com/models/774290?modelVersionId=866020</a><a target="_blank" rel="ugc" href="https://civarchive.com/models/775853?modelVersionId=866020">https://civarchive.com/models/775853?modelVersionId=866020</a><a target="_blank" rel="ugc" href="https://civarchive.com/models/775853?modelVersionId=866020">https://civarchive.com/models/775853?modelVersionId=866020</a>I recommend you use this at 0.8, but you're free to experiment. Use the tags as provided for the model, or experiment if you want to. I'm not 100% sure of how compatible they are to be used together, don't try using them all at the same time, it will probably not work. Also, the anima model's recommended weight is 0.7, but I recommend making it lower if you wanna use it without a specific style in mind.

Drowned Sorrow (OC collection) [Illustrious/Anima]

Marisa's anima version. Due to some mistakes I made in training, I recommend you use the lora at low weights, around 0.3 to 0.4. The reason for this is that apparently her mermaid related tags (mermaid, mermaid tail, red tail, head fins) got tagged twice in the images by mistake. As such, these tags have recieved significantly more training than other tags to the point of overtraining, which means that unfortunately a lot of the dataset's general style faults (which I was actively trying to train away this time around, ironically enough) are a lot more pronounced compared to other characters using a similar dataset. This is something I very much intend to fix before the big version merge later.

Files with SHA256: 428670ee67b2812a099a28c0013cd10b341db86f7639a5f29723796facd0cb28