A LoRA for phazon, phazon corruption, and phazon tentacles. Comes with a side of Dark Samus because of bias in the training data.
Anima_v1: (in upload) Trained on Anima base 1.0 using the same training data as the ILv1.0 versions. Triggers:
Base: phazon, blue glow, slime \(substance\)
Tentacle: phazon tentacles,
Sime girl: phazon slime girl, slime girl,
Corrupted by phazon: phazon corruption, glowing veins,
Has some weird style effects, so you may need to add "3D" to the negatives and/or add "digital art" to the positive.
ILv1.0 - v1: Trained on Illustrious v1.0 with slightly expanded training data and better quality tagging for significantly less impact on eyes and somewhat less impact on hands.
Main trigger word: phazon, blue glow
Drops "blue glow" to get other colors, but it is required to get the traditional phazon look.
Additional triggers: phazon corruption, phazon tentacles, phazon slime girl, slime \(substance\). These can be used piecemeal. The effects aren't exactly the same as they were for the previous version, but they're close.
Noob vP1 - v1: Trained on NoobAI vPred-1.0, same data and tags as ILv1.0 -v1.
IL0.1 - v1: First version for Illustrious 0.1, also works somewhat on ILv1.0 and NoobAI based checkpoints.
If requested by Nintendo or another Metroid Prime rights holder, I will remove this resource.
Description
First version for Illustrious 0.1. May work with NoobAI and Illustrious 1.0.
FAQ
Comments (6)
Magnificent lora, do you plan to do more with the concept?
I don't have anything in mind at the moment. I can't run any of the post SDXL models locally, so I don't intend to do a Flux version or anything like that. I could try to do a NoobAI or Illustrious 1.0 version, if the current version doesn't work with those, but that's about it. I haven't found much, if any, more training data, so a v2 isn't likely anytime soon either.
@magnoveradium Got it!
If it would be helpful to have a massive dump of images generated with it, I could always provide that (considering the potential for a v2), if you'd like. I stick pretty much entirely with illustrious based models, yours works okay with illustrious v1 but the textures can get a little fucky sometimes and it can influence style a bit. I'd be happy to contribute 1k buzz toward training on a v2 or one trained off of the ideal current Illustrious SDXL model (I mostly use perfectdeliberate_XL at the moment, but it's really your call as to whether that's the ideal to train on or not)
Either way, Phazon or otherwise, I dig the whole 'transformative slime' concept, and would be super interested in a model of this style/quality that broadens it beyond Phazon-specific concpets as well.
@Shiggles I noticed the style impact when I made this, it's probably related to how quality and style tags work in Illustrious vs how I tagged the images for training.
I'll have to look into training vs IL v1.0. I'm hesitant to train on images made with this LoRA because of the potential to reinforce the style impact.
I think there are loras out there for slime transformation. I'll have to check how those work before spending the time to collect art to train one myself.
@magnoveradium Very fair! And yeah, style impact can be hard to escape with limited datasets like this. My three character LoRA's are clearly victims of this. Either way, I'd be happy to fund a thousand buzz if you find yourself looking to train similar concepts, just hit me up!
@magnoveradium I see you updated the model! Thanks a bunch, you're the best.