* Re-train version is trained on cleaner versions of the original lora's dataset images and finally removed the annoying halo/antennae. Works much better than V1.
<lora:Darla2:1>
* Version 1 works well but the re-train is a lot better.
<lora:DarlaD:1>
(Only for V1): Add "antennae" and/or "halo" to neg prompt if necessary.