Ghostboy (Patrick Fillion charecter)
Details
| Model Type | LORA |
| Base Model | Illustrious |
| Published | 8/6/2025 |
| Trained Words | pf_ghostboy Copied! white superhero mask Copied! white superhero top Copied! white superhero leggings Copied! |
Download Files
About this version
V3.0 retrained for Illustrious with an update dataset to try to get the mask appearing correctly (or at least closer to what it should be) as well as continuing to try and have the chest tattoo.
trigger word: pf_ghostboy
other keywords:
white superhero mask: self-explanatory, all dataset images where he's wearing the mask contain this keyword; additional related keywords that may be useful
black triangular eye markings: an attempt to govern the triangular pattern around his eyes.
exposed brown hair: to indicate the mask doesn't cover his hair
glowing red eyes/pinpoint red eyes: an attempt to govern the eyes. pinpoint red eyes was used in the dataset where images only showed a red dot, whereas glowing red eyes was used for images where there was a discernable iris or outright glowing eyes. not sure if this was as successful as I would have liked.
abstract mouth design: an attempt to govern the mouth shape/design, if this doesn't work I'm had success with jack o'lantern smile drawn on mask, but that keyphrase isn't in the dataset so it depends on your checkpoint. Probably use that for inpainting if you get desperate (as opposed to the initial generation prompt)
skintight full head mask: alternate prompt for white superhero mask in case you're using a checkpoint that tends to splash superman's/batman crest on anything marked 'superhero'. any image in the dataset that has white superhero mask also had this tag, so you shouldn't need to use both,
white superhero top: all dataset images where he's wearing the top half of his uniform were tagged with this; additional keywords that may be useful
white skintight crop top: the alternate for white superhero top in case your checkpoint heavily associates superhero with established heroes (i.e superman). any image in the dataset that had white superhero top also has this tag so you shouldn't need to use both.
full sleeves: shouldn't need this, but it's in the dataset if your checkpoint keeps giving him short sleeves
white gloves: shouldn't need this, but it's in the dataset if your checkpoint keeps giving him bare hands
white superhero leggings: all dataset images where he's wearing the bottom half of his uniform were tagged with this; additional keywords that may be useful
white skintight leggings: the alternate for white superhero leggings in case your checkpoint heavily associates superhero with established heroes (i.e superman). any image in the dataset that had white superhero leggings also has this tag so you shouldn't need to use both.
pulled down leggings: this was included in addition to white superhero leggings in all dataset images where he had pants on, but they were either pulled down or tucked under his cock/balls. You can probably use any other similar phase you like and most checkpoints that can do NSFW will handle it just fine, but it's in the dataset
Model description
Made this for personal use and also as part of a tutorial on creating LoRA, as such this is very alpha build, (let's be honest I'm not quite sure what I was doing)
Regardless, uploaded so you can have some fun with it for personal use.
read the version notes ->











