Hi ヾ ^_^ this is my first time making a Lora !! so I wanted to test it with a cosplayer I love a lot https://www.instagram.com/ri.ggwp/. I originally made this because I wanted a face that I thought looked a lot like Overwatch mercy but in real life and I think she fits that!
I HIGHLY recommend using https://civarchive.com/models/10961/lazymix-real-amateur-nudes as the model or a model similar. Chillout mix's do not work very well!! (just a heads up) I will try in the future to make my loras more universal my apologies
There are Two different version I recommend ri.ggwp_2 first and if you don't like it try my fav version ri.ggwp_1 !!!
This Lora works best at high weights 1-0.75 lower weights tends to loose her likeness. I didn't use any of her cosplays in the data set i only used photos of her out of cosplay as a heads up !
I also highly recommend putting the hair color you want her to have since I made that very flexible.
If anyone wants to recommend any other cosplayer i would be more then happy to make a Lora of them ^ω^.
Pls be kind with this lora <3
Description
This is just a higher epoch version if that's your taste !
512x512
1.4 upscale
1-75 weight like the first version ^_^
FAQ
Comments (7)
Hi! Which checkpoint did you use as base to train this? I've tried chillout but it tends to ruin the skin details.
yeahh I was literally just testing it with chillout and it doesn't work at all. I highly recommend https://civitai.com/models/10961/lazymix-real-amateur-nudes I'm using a mix of this and LOFI. I would recommend if your seeing messy/greasy skin details to lower the weight just a little bit! my apology's if it isn't amazing its my first Lora so I'm hoping to improve in the future (〃＞＿＜;〃)
@Elluna Actually I've tested this LoRA with multiple other LoRA and it works pretty well. Mind if I ask how you set up those train parameters? I'm not sure what I did wrong but my LoRA produces glitched face, and I can't seem to find a decent tutorial which covers everything. Of course if you don't feel like sharing you don't have to. Thanks btw for this LoRA. Really cute😊.
@Elluna Ohh and it does work well on chillout, you just need to install this extension, and use it like this:
<lora:riGgwpLora_riGgwp1:1:KEEPFACE>, where KEEPFACE is a block weight preset I tweaked to only keep the face:
KEEPFACE:1,1,1,1,1,1,1,1,0,0,0,1,1,1,1,1,1
@NotAPersonJustACat OMG I didn’t know that extension existed I’m downloading that immediately thank you (∩˃o˂∩) !!!
I used this guide https://civitai.com/models/22530 plus some random Lora training videos to figure everything thing out. I use the google colabs listed there as I find it easier to do it in there then in SD GUI. (I also have a google collab subscriptions since I couldn’t train more then one Lora with the free time). Anyway Here’s the config info for my lora:
learning_rate = 0.0005
lr_scheduler = "cosine_with_restarts"
lr_scheduler_num_cycles = 3
lr_warmup_steps = 108
If you want more advice for training such as links and just my own advice if you have a Twitter pls message me there (linked on my profile) I can help you through dms and send photos of what I mean and such!!! Hope this helps and thank you so much for testing my model further for me (づ๑•ᴗ•๑)づ♡ !!!
@Elluna Thanks! I'm going through that tutorial rn. It's quite different from training locally :( I hope this time I can get it right.
@NotAPersonJustACat I believe in you and if you ever need assistance I’d be happy help (๑ᵔ⤙ᵔ๑)