This merge aims to keep the vibrant colors and contrast from NoobVpred models with minimal anatomical inconsistencies while preserving the compatibility with LoRAs trained on Noob or Illustrious models.
⚠️ IMPORTANT NOTICE ⚠️
THIS MODEL WORKS DIFFERENTLY THAN EPS MODELS!
Implemented as a v-prediction model (distinct from eps-prediction), it requires specific parameter configurations - if you get black boxes you probably need a more recent version of A1111/Forge/ReForge.
For more details, you can check the info on the Noob page.
Recommended settings :
Sampler: Euler / Euler a
Steps: 25-30
CFG scale: 5
Merge from :
- NOOBAI XL-VPred 1.0
- NoobAI-XL-Vpred-v1.0+v29b-v2
- ΣΙΗ
- WAI-NSFW-illustrious-SDXL v12.0
Description
FAQ
Comments (16)
EDIT: Updated Forge as recommended. All working now!
Downloaded this three times now but the model hash doesn't match and it just produces empty black boxes. Think the issue must be on Civit's end but maybe someone else can confirm I'm not insane lol?
I will check and see if I can reproduce the same error.
I tried it and it's all good on my side.
On Forge: Same hash (675dd999c6), same pictures generated.
@Konan Super weird, I've tried multiple browsers now and downloaded several different times and it's just churning out black boxes and the hash in forge is b3bf245f86. Other models working normally. I'm assuming it's GOT to be a civit issue then.
@ArtificialAnaleptic so I got someone that replicate your error. They are using A1111. I will investigate.
@ArtificialAnaleptic Ok you probably need a more recent version of A1111/Forge/Reforge to run it because it's a V-Pred model. I was able to get black boxes using an old A1111 and not only with my model but on other Noob V-Pred models.
@Konan Thank you! I'll take a looks and see if that fixes it.
@Konan All working now! Updated the OP comment to recommend an update if it's not working.
I was trying it out before the revert, and it's actually pretty easy to use. The colors are still great, and it's far superior to several checkpoints I managed to try this week... Gosh, I guess I'll try using it locally. XD
I had big hope for the generator :(
Thanks for the feedback
takes less prompt wrangling for consistent contrast/temps/lighting than some eps ive used, wth
also very good for inpainting which is another surprise for a vpred model (for me, im no expert)
really really good
What does the negative tag "4koma" do?
Exclude panel/multiple views and badly draw pictures
https://danbooru.donmai.us/posts?tags=4koma
i get hella saturated colors, is there anyway to turn this down a little bit ?
Cfg = 5
Don't go above.
Easily the best model out there to create classical paintings or digital art in Illustrious without having an anime styled face or artstyle