UPDATE: If you wanna run this model on Noob, or Noob-derivative models with Civit online generator, make sure you turn on advanced settings when selecting a model.
The Noob version seems to kinda work on the Civitai generator if you drop the CFG scale to ~3.0 (I'm using Euler a). It's not gonna have the same quality as genning locally since Civit doesn't support Vpred yet. Using the Civitai generator will result in oversaturated images.
A custom style based off of a bunch of artists/mix from NTR Mix, which I generated a bunch of images off of and trained for NoobAI Vpred.
This lora is trained using Vpred paramaters so please ensure you can run Vpred models. It works best on NoobAI Vpred v1.0, although it might not work on the Civitai generator because they don't support Vpred (yet).
Recommended lora strength 0.7-1.0 (I usually use 0.8-0.9). Don't forget to use "Furry" at the beginning of your prompt for better Anthros.
Description
FAQ
Comments (21)
You set the basic model as Illustrious when uploading, so it can't be used with noobai on site. Can you change it to sdxl?
Are you trying to run it on Civit? That won't work well because Civit doesn't have Vpred support yet (I think)
@KumquatMcGee it works, thank you
i cant seem to recreate similiar images, is there any hidden parameters not included on the metadata?
There shouldn't be. Are you genning locally and have all the Vpred options enabled?
@KumquatMcGee I'm using it on reforge, but the same problem happens with forge. i enabled CFGrescale, i read somewhere that i only need to enable this. i can recreate the characters but not the general composition, like the characters would be on opposite sides and have different expressions and what not
@Daqenya Try playing around with your sampler and CFG settings. I am using Euler a with 6 CFG. I know it works because I've seen other people download this model and it works fine.
I'm having a problem with the lora, for some reason it ends up creating a very strong orange saturation, I'm hosting locally with an 8Gb gpu, and the model simply doesn't create anything relatively similar to the images, and in 20 steps they look horrible I'm using forge limited to 7000mbs to avoid the low vram bug
Are you using ReForge and you have Rescale CFG enabled? You need that for Noob to begin with
@KumquatMcGee no, i'm using forge, not reforge, i'd have to change ui to use it ? i'm using a lower cfg for now, it's improved slightly but there's still a strong orange and yellow tint to most of the images
@hexmachina Sadly yes, Vpred XL models like Noob only work with ReForge IIRC.
@KumquatMcGee ok i'll use the reforge, as soon as i'm done i'll see if anything has changed in it, if so i'll come back with positive feedback
@KumquatMcGee ok, how do you use cfg Rescale i don't understand how it works and i have no idea how to use it, is there a good preset ?
@hexmachina once ReForge is installed, just click on this setting: https://files.catbox.moe/z73lpf.png
and it should work. I am using Euler a @ 7.0 CFG
!!Just as a FYI for any base Forge web ui users!! that are getting oversaturated imgs even when using "CFG Rescale"; that's because CFG Rescale doesn't work with base forge BUT, there is an integrated version that comes with base Forge that does work, so everyone should have this.
Once you've launched Forge, you'll want to scroll down and find a tab called "LatentModifier Integrated", click on that to open the tab, select "Enable", then scroll down till u see a setting called "Rescale Cfg Phi" and set that to the value u want; I typically use a value of "0.7".
Hope this helps out any base Forge users like myself.
Yeah that or use ReForge https://github.com/Panchovix/stable-diffusion-webui-reForge
Can you tell me what Vpred parameters are needed to train the vpred model? The lora images I trained with noobaivpred10 are all gray and can't be seen clearly.I am almost desperate.
I used this: https://imgur.com/a/uv9xdSj
Here is my TOML for settings which has everything enabled: https://files.catbox.moe/pzfufa.toml
Finally i found you. I've seen your art style and have been searching all over for the source of it. And now i finally found it.
Your lora is amazing. You've inspired me to get into AI generating myself. Thx for sharing it :)
Thanks, where did you see other pics generated with this style?
@KumquatMcGee They made their way to the Rule 34 websites. After searching among those, i found that some of them had a links leading to some desuarchive threads full of anonymous users sharing AI generated artworks.
After searching for a while, i found the same images uploaded to R34, and an anonymous users had posted a link leading to this profile. And that's how i found you.