Even just a like would be an awesome help to boost this model! Please post your results if you like my model as well!
(Following applies to v2.0)
All-in-one, UnOrdinary Style + Character + Ability Lora
Covers 25+ characters.
Does not need additional LoRA at all. Will give very good results with just base model + good prompting.
Functions well at high strength, with very dynamic style control.
Over 210+ image data set.
Best Recommended Settings:
CFG Scale, 3 to 7
CLIP Skip, 1 or 2
Strength, <= 0.9
Primarily designed to be a "complete" All-in-One UruChan / UnOrdinary Style LoRA that should not need refined or added onto. Focuses Season 2 Part 2, and early Season 3 art, along with the best more complete Instagram works from Uru. 95% of tags are identical to Danbooru tags, so it should respond well to them if you're looking for angles or poses.
This is primarily designed to be a Style LoRA, retention of style is prioritized over pure canon accuracy, though it should be good at both (At least in the Heavily Trained+ characters), because of this character features like hair and eye color are not tagged, you should get better results when prompting non-canons with hair and eye color tags because of this, as they shouldn't overfit details of canons, you however likely will get less sharp features among canons, simply using more descriptive tags similar to canons for their hair shape and such/details should fix this for most characters (Particularly those in the Lightly Trained category).
I manually removed ALL Chat Bubbles, you should not get Chat Bubbles at all without any prompting.
Style Tags
/uruchan style/ (Main Trigger Tag.)
Character Tags
Very Heavily Trained, John, Arlo, Seraphina, Remi, Blyke, Kuyo, Isen, Valerie, Kassandra, Leilah
Heavily Trained, Elaine, William, Farrah, Cecile, Leilah
Moderately Trained, Vaughn, Jane, Cameron, Rein
Lightly Trained, Zeke, Keon, Claire
Very Lightly Trained, Rei, Kayden, Darren, Grayson, Liam, Tarik, Joker
Cool Extra Tags
Glowing Eyes (Ability active glow of eyes)
Glowing Body (Glowing body)
Trained Negative/Additive Prompts
Speed Lines (Can manually add or remove them with this).
No Eyes (Far shots/Full body will look much more stylistically consistent to Uruchan with this tag, you can negative tag this if it gives issues/you don't want this).
Score tags (Score_9, Score_8...)
Best Quality
Blood
Scars
Description
Final Version for Illustrious (I will be moving to Anima checkpoints for future models. I've found Anima so far to be vastly superior for the kind of models I've made.)
Dataset Pruning
> About 35 images were removed, 54 were added. Adding 19 images overall to training data. I want this LoRA to be at it's absolute best. The primary purpose of this version was to finish off at least getting most characters reasonable datasets, characters like Rein and Cecile were not integrated enough to generate. I wanted to fix that. And more majorly I noticed my LoRA struggled to create full body images, I cut back on some of the worse far shots and added a few good ones. All images were entirely retagged from scratch with the new knowledge I've learned over the last 8 months.
Tagging Overhaul
> /uruchan style/ is the main trigger tag now
> Tags were simplified, all characters now tag as uno[insert char] (Example: unokassandra). You can mostly thank John for this, I think most of John's generation problems were because he was just tagged as john.
> ability/... was removed. Instead now you should mostly just be able to simply describe the actual ability (Example: red particle spikes). The old formatting simply wasn't the most effective way to go about what I was trying to make. Clothing tags should be better also now, I specified them a touch. General description tags should work better now.
> Score tags have been added. This is mostly because this dataset was designed primarily for anima now, but you can use score tags now, adding score_9, score_8 will tag the top ~35 images in the dataset, and negative tagging low scores similarly will filter against backed images.
Character/ability additions
> Isen - Predator (and him generally) still wasn't tagging well. Many more images of this and him were added again. He now has MORE images than Blyke and Kassandra. Tag tiger stripes and predator form for his new(ish) predator ability.
> Valerie - Previously was slightly short on the amount of images need to accurately make her. It was serviceable before, should be great now. Her dataset is about as large as Farrah's now.
> Leilah - Could not appropriately generate Leilah much at all, varied her dataset more, should generate about as good as Valerie does in this new model.
> Darren - Probably still so-so. He basically had no images before though.
> Cecile - Couldn't generate well before, dataset is much larger for her. Similar to Valerie now.
> Elaine - See above, she got the same treatment as Leilah and Cecile
> John - Almost all images were pruned to his current hairstyle (He was the big pruning and reworking here). It simply could not work well with his hairstyles varied too much, hopefully you can get John-John instead of John-Slightly-Off-Hairstyle-John now.
> Blyke - Blyke got some minor changes to get him more accurate to his current hairstyle instead of early hairstyle as much.
> Claire - Claire was not in the dataset at all before. She got a few images added to the dataset, I wouldn't be all too confident in it, but she is tagged and added here now
> Rein - Rein was a soft spot for the previous LoRA due to bad image quality, I upped the quality of her images and added a few more.