license: creativeml-openrail-m
tags:
- stable-diffusion
- text-to-image
datasets: embed/EasyNegative
Descriptions
This model is suitable for reproducing images similar to anime screenshots, and it is recommended to use LoRA block weight expansion when using LoRA and character LoRA together. Setting the block weights of the character LoRA to presetting MIDD will reduce the distortion caused by the influence of the character LoRA on the style.
It'll give you clean, good anatomy, and very anime-like pictures with beautiful backgrounds!
There are two models provided, a LoRA model (which is only 36MB!) and a full base model (which is literally the LoRA model merged with Anything V4.5). The full model is provided for training your character model. I found it's good for training anime characters with this model.
Recommend settings:
- VAE: Orangemix（the same with NAI）
- LoRA Strength: 1
- Sampler: DPM++ 2M Karras
- Sampling steps: 20
- Negative embedding: EasyNegative、badhandv4
Highres fix is also recommended.
Samples
{% for img in 'img_list.txt' | readlines %}  }}) {% endfor %}
Models used
Merged with block weights tweaked:
- 2020s Anime Magazine Illustration Style
- Anime-like 2D (extracted LoRA)
- Anime Lineart Style
- Anime Screencap Style
- Avas Anime Hamster
- Epi Noise Offset
- Hipoly 3D Model Lora
- Makoto Shinkai Substyles
See also
Original post on Civitai: https://civitai.com/models/23723