Trained on 65 curated images.
Use 0.7 - 1 for weights.
I'm including casual prompts just in case you want to try them out, but you can just ignore them, the purpose of this Lora was not focused on Birdy's normal looks.
This Lora is mainly focused on her battle suit as well as her appearance. You can play a little bit with the prompts and somehow get her casual brown hair as seen in some examples.
As a note, you can sorta play with the color of her hair, it's mostly pink but changing pink for orange also works, sadly I didn't include any example.
For better results as always use Adetailer and Hiresfix, examples were generated in AutismMixConfetti.
Description
Same exact data and configuration from the Pony version, it gets the details way better IMO, still, if anyone can throw suggestions for training configs (using the trainer from google colab) I'll be happy to test them.
There is a slight problem if you want to prompt her with the brown hair, seems pink hair is too strong so I suggest you to use a strength of 0.7 - 0.8 to avoid this.
Trigger words are the same and if you're using Civitai helper you only need to click the lora card and everything should be added automatically! Unlike those LORAs with insane amounts of tags all spread into different images! Also the size of this lora remains the same, none of that 217MB pesky loras! We like our LORAs small, no one wants to waste 1 GB to 5 of these! No no no, 54MB should be the bare minimum!
I don't use regular configs for testing but everything was working okay as you can see in the metadata, I imagine all of you will be able to use this LORA way better in more creative ways!
FAQ
Comments (1)
As far as getting brown hair, the reason orange works is because what we associate with brown as pixels emits a very dark orange. So specify dark orange and it should appear to be roughly brunette to auburn.
However you may require some additional prompts, LORAs, negative prompts, and whatever other else necessary to reinforce it and get exact results.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.