SteelyBird Art Style DoRA
Details
Download Files
About this version
Model description
Please support SteelyBird by visiting their socials like Twitter, Patreon or Subscribestar. This kind of thing wouldn't be possible without the original artist and their amazing work. If SteelyBird, or a representative of theirs contacts me and asks me to take the model down, I'll respect their wishes.
The purpose of this DoRA is to try and capture just a fraction of a fraction of what makes SteelyBird's art so great. Given that this is my first attempt at training a LoRA/DoRA, I'm pretty happy with the results. I'm happy to hear any feedback from people in the comments/my chat to learn how I can improve it. Please share your gen. results, and share prompt info so I can find out what works best! BTW, big thanks to bluvoll and THEANTLERS for their help and advice with the creation process.
Below is a list of the good and bad of this model. I think the bad can be fixed with tagging adjustments.
The good:
This model will not produce any watermarks, as those were all removed in the dataset preparation process. Speech bubbles were scarcely present in the dataset, in the case that they were impossible to remove without ruining the image itself. However, they were tagged for, and so they won't appear unless you include a tag in your prompt.
If you like 'em thicc, boy have I got a DoRA for you... generally the "large" tag is on the small end, "huge" is medium, and "gigantic" is absurdly huge when used in combination with a part of the body. It shouldn't be necessary to increase the weight of any body size
This model seems to succeed in capturing certain traits of SteelyBird's signature art style, such as the proportions, faces, simple sketch backgrounds with detailed characters, and soft stuff being smooshed.
The bad:
Pink hair seems to always come along with a magenta hairband, which is a trait of the character Wendy who is prominent within the dataset.
Thighhighs may default to looking like 2b's thighhighs due to her prevalence within the dataset. This is something I could go back and adjust rather easily.
Glasses seem to be really wonky. This may mean that a tag adjustment is necessary.