Dragon maker. Unlike the pony version i dont think the lora can do something Illustrious base cant do, but i was curious ;).
Illustrious :
v1 :
Trained on illustriousXL_v01
Weight : 1 - 0.8
Trigger : dragon, teeth, sharp teeth, open mouth, wings, monster, scales, horns, claws, tail, spikes
Would be awesome if u leave a like and post a pic ;)
Description
FAQ
Comments (3)
"i dont think the lora can do something Illustrious base cant do"
Honestly, I'm mainly downloading this in hopes that it produces better teeth on my dragon-style characters. Although a part of the issue there is possibly that I always go for realism/semi-realism, and all images of dragons are not real.
So, there's a bit of a gap in turning it real in the first place, and a lot of the training material for dragons will have bad angles(cuz angles are really difficult to master in art, even for many pros), and flat color images which don't give the AI a good map to go off of for how to make the teeth look in 3D.
If you don't do it, I may eventually do it myself, but the ideal method for making a great dragon lora, would be to tweak CFG, steps, lora/clip strengths all until you can generate the best 3D, then generate a whole bunch, and use the best of those to train another lora.
That is after all, the true strength of AI, not copying us, but improving on the input that we feed it.
Also, if generating for training data, unless you have a limit to what you can generate (IE: you can't afford an AI PC), then discard any with the slightest flaws. Be very picky, cuz AI will quite often give us a real serious gem of an image from every 20-100 or so, and if you only use those, then you can make the best loras.
Additionally, I recommend, well, actually, more of a request but please don't bias with human female data at all. The main models are seriously oversaturated with human female images (anime, and realism), adding any more to loras arbitrarily will only make the lora push female features even when people are attempting to generate male characters, or no humans at all.
And if a lora is for personal use only, then there is no problem with that at all, but if uploading into a space where all types of people will use, then it is best not to overly push personal bias into the loras when at all possible.
Strangely your comment wasnt shown in my notifications. All dataset images were without gender indication. just dragons from different games and artworks. and it was just for fun, if i ever need something like that for an image i want to make. and im curious how something like this is interpreted b ythe AI. did u make images with it? was it useful in anyway?
@BloodySunkist sadly, I don't think I have, yet. I haven't been as into image generation lately, been spending the bulk of my time on my PS5 or tweaking/adding new features to a fully local AI/model reader program(views/displays metadata from Loras/models).
I plan to release it on GitHub soon. And will prob add it on here as well just to get it out there.







