Home Artists Posts Import Register

Content

As I was anticipating in the release post for Mistoon_Pearl, this is not a beginner-friendly model. When working with this model there are a few things you'll need to consider:

  • This model works better when used with a low CFG value. I tend to use 4/5 when generating txt2img. In img2img you can go higher (around 7).
  • All the pictures from the previews have been created using one of these two schedulers: Euler A or Restart.
  • Having at least 30 steps can highly improve the quality of the generated picture.
  • The model does not have a consistent style by design. It has been trained with pictures with different styles with the purpose of having a higher variety than the standard Mistoon models.
  • From my testing it seems that the model works better with pictures with a 2:1 ratio. I usually start with a width: height of 384x768.
  • This model works really well with img2img and IP-Adapters

Let me know if you find any other useful tips!

Files

Comments

Anonymous

Is it friendly with lora? I keep getting some pretty bad results, im following the suggestions too

Anonymous

can you provide an example prompt? So i can test as the baseline

Inzaniak

Unfortunately, right now it's not LoRA friendly. Only a few LoRAs I've tried gave satisfactory results. I'm trying a few things to improve this, but right now I suggest to avoid using LoRAs with it.

Inzaniak

The example above has been created using: Positive: question mark,confused,white collar,white shirt,bookmark,eris kissaten,mushoku tensei,blue hair,from side,braid,collar,long bangs,long hair,upper body,looking at viewer,shirt,blue eyes,rolled up paper,ahoge,roxy migurdia Negative: badpic Seed: 4224346749