Diives- Style LoRA

INTRODUCTION

800 Images used for training check Diives out https://www.patreon.com/diives

In short, this LoRA does it's best to replicate Diive's style, if you don't know who he is (which...if you're even remotely on the internet how can you not? Especially in this avenue of the net) look him up. You'll probably be there for a while because his discography is pretty sizable. Recommendations will be below. If I were to describe it, his style is semi retro mixed with a stylistic flare only he has, but not anymore MWAHAHAHAHA...I hope. This is my first LoRA creation so this probably just sucks without me realizing, but that's what the comment section is for, go ahead and degrade me all you want as long as there's at least some information I can use to try and potentially make this LoRA better. Made using https://github.com/hollowstrawberry/kohya-colab

BEFORE

This LoRA can be nsfw or not, I'd attest to nsfw due to the fact many of the free Diive videos are censored so the ai may struggle in coming up with...how should I put this...genitalia structure, work with boobies just fine most of the time.

The whole furry thing has been a mystery to me, I'm not sure how to get the ai to identify furry prompts consistently even through a few hours of testing. The closest I've gotten is the ears and tails.

Feel free to experiment and tell what has worked for you in the comments, this is my first LoRA so I can only get better. I know all of you are sweaty nerds anyway.

The ai will sometime fight when it comes to positions though this again is an inconsistency, for example if you tell a character to stand in the prompt the output will probably not have the character standing

It has potential to work with other LoRAs yes, I haven't tested it with any game characters however.

Please don't be a bitch and use this LoRA like a menace to society, credit to Dave for his amazing work. With that being said do not distribute this model or refuse to credit Me and Diives, actually just Diives he's the one that did all the work with all the artwork. Anyway let's get into the recommendations.

Recommendations (I would prefer you do some experimenting, this is just a starting place)

• Sampling Method: DPM++SDMKerras But I recommend you experiment, see what works. Be sure to help out and tell what you found worked in the comments, I'm just ball parking it as of late.

• Sampling Steps 30

• Hiresfix on, steps: 15

• I have to use 512/512 because if I don't my 8gb of ram prehistoric laptop will explode, I'd think using the anime upscale will work fine but prove me wrong.

• Denoising Strength: 0.7

• CFG scale: 9

• ADetailer(1): face_yolov8n ADetailer(2): hand_yolov8n

• Inpainting: Restore faces after ADetailer

• For the weight I recommend starting with 0.7, you'll know when the weights should be moved when the generated image is a bit too bright and or distorted. Shouldn't be moved past 0.5~0.8 from I can see

  Keep in mind that this LoRA was tested with this model it may not work with others, it was also trained with the same checkpoint.

  check Diives out https://www.patreon.com/diives
  That's it from me, peace!