for usual appearance try any of these or put in Neg (changing appearance/outfit)
may need to add 3d, or realistic, in neg for certain 2d/anime styles due to having more 3d image training or lower weight for less 3d influence
ILL - FleshPits was added has less general training and has no default attire training
v2 FleshPits has pod training, though it is tricky to prompt and may need use of NEG such as fish, robots, etc
ILL - DF V2 may have less 3d strength than V1
For appearance
sd1.5 -> Mileena, black hair, medium hair,
yellow eyes, split pupils, red sclera, or glowing eyes
extra teeth, or sharp teeth, if no mask or put in neg for less horror
may need to put lips, nose, teeth, prompts in NEG if using mask
1.5 - tongue prompts, dont work as well as the others could try support loras for better results
for outfit
pink or purple leotard, cleavage, or center opening, fingerless elbow gloves, thighhighs, or thigh boots,
suggest covered navel, in NEG or add navel in prompt
may have to specify the pink - put yellow attire, in neg
types of mask like cloth mask, etc
My other mileena loras
mk 10 - https://civarchive.com/models/111851?modelVersionId=392693
mk1 - https://civarchive.com/models/152643?modelVersionId=210119
Description
FAQ
Comments (6)
It seems to be a good LoRA, the problem is learning how to use it. I did some tests, and all the images came out noisy. The biggest problems were generating the eyes and teeth. It would be good if, in the example images you use, you included the original LoRAs used in conjunction.
I'm not sure what you mean, I test them a lot before putting them up. Other folks do showcases just fine if not better than my examples.I'm not doing anything special with the lora. I don't use more than one lora at the same time in almost all my showcases.
What you see is what was used without any other loras involved. Except for some of the models/checkpoints not available here in my example that uses the lora itself. I don't use textual inversions or lora enhancers almost never
It sound like you need to use adetailer for the face and eyes like everyone else. And if that's not reason then ???
Dude, I really like your LoRAs, the Mileena from MK10 is a masterpiece and I can use it perfectly. What I did here was constructive criticism; I'm not saying LoRA is bad, I'm just saying it can be a bit inflexible with other LoRA styles, and that (sometimes) eyes and teeth get distorted in full-body images. And about other people making demonstrations as good as yours, well, in this specific LoRA version there are only 3 posts so there's not much to work with. Don't take it personally, bro.
@Surprisecondary_account its not personal, because I didn't say anything about you as a person. Its just not working well with 20 other style loras is not my issue. if it does "cool np" but I dont make my loras based off on that. I had to make many style loras for myself because folks are too lazy to upscale images or remove watermarks or make changes to them in photoshop for datasets.
And by other peoples showcases I mean all my recent ILL loras updates not just this one. I make my showcases simple on purpose. "you included the original LoRAs used in conjunction." <-- My answer is that I didn't use any other loras. if the lora needs other loras to work? what kind of character lora is it?
People always having a stack of enhancers for a single picture and be like 0_0! when things break has nothing to do with the character lora by itself. You said it was "perfect" which tells me the lora is good enough when used alone. Could it be improved? of course but her face being distorted because someone's else style butcher faces when used with some other style? is not my problem.
Your OP sounded like you were asking for some sort of guide or explanation regarding noise and face. Then you told me what the issues was which was on your end.
@True_Might I retested your LoRA and it seems to work better with lower weights. I'm used to using LoRAs for characters with a weight of 1.15, and it seems that above that weight things start to get "strange".
Sorry for the inconvenience, but here at Civitai there will always be people with doubts and questions (like me). I don't understand why some creators get defensive or take things personally when they receive something that isn't praise.
Just to end this conversation, I apologize again, and I repeat that I'm not criticizing your LoRA, it was just a small problem finding the right weight, please don't be angry.
@Surprisecondary_account its all good, I was more confused if anything,
I will say that my loras tend to have a lot of 3d strength so lowering the weight helps it with style loras. And of course the usual 3d/realistic in NEG.
Im aware they can struggle with art styles because of the nature of the 3d model / realistic graphics images I use in my datasets for the lora. I try to add as much 2d fan-art as possible that dont alter the character too much so its easier to do what your doing. So its not like I dont want people to mix things together. I merge style loras into checkpoints instead so that why it seems like Im using several loras at once with the character lora for a prompt.
I know the loras are not perfect because im always randomly updating them. Is the ILL v1 version of mileena still better than v2? Im not sure. Regardless of how other people showcases look great I usually update the lora anyway.