Three Delicacy Wonton (三餡馄饨Mix) – V1.3+Anime

Three Delicacy Wonton (三餡馄饨Mix) - V1.3+Anime

Run Model 使用GRAVITI Diffus 关于GRAVITI Diffus
Model Name Three Delicacy Wonton (三餡馄饨Mix)
Model Type Checkpoint
Base Model SD 1.5
Version Name V1.3+Anime
File Name threeDelicacyWonton_v13Anime.safetensors
SHA256 2ecd09417d
NSFW False
Trigger Words
Tags anime, watercolor, chinese, style, hanfu, illustration, artstyle, gyoza-based
Metadata fp = fp16
size = full
format = SafeTensor
Size 3.95 GB
Version Description

我…我菜谱普掉了。 I lost the recipe ;_;

I think it was 1.0 KimHongDo, 0.7? Shukezouma, 0.5? Xiaorenshu, 0.5? MoXin, 0.8 watersmudge, 1.0 A1 filter? I recreated it and it seemed to be largely the same.

Author Jemnite


  1. 切一点金弘道 (

  2. 拌一点墨心还有疏可走马 (

  3. 刷一小点小人书, 就一小点给配风景 (

  4. 都包在一个大皮子里边 (


Huggingface repository

Ashen’s ColorBox color-enhancement if you like the style but wish it had more color.


A LoRA mixed model of my earlier 餃子Mix. This checkpoint kind of puts the whole picture through a monochromatic filter reminiscent of the 水墨畫 style calligraphy used in Sinosphere traditional paintings. It has an extremely heavy effect on trees and landscapes, where the leaves turn into the ink blotting reminiscent of calligraphy painting. Less-so on the characters, as a result the tuning ratio has been adjusted to preserve the anime-esque style of the characters.

Originally I was going to name it 虾饺Mix because I love shrimp jiaozi, but since I added a bit of the 伢伢書 LoRA, I figured I’d make it more Shanghai-esque (plus I had wontons for lunch yesterday). Hence, 三餡馄饨Mix.


As of now there are 3 versions. Base, Anime, and MSG. Base is just GyozaMix with a light filter. It’s okay to take in LoRAs. Season to your taste. Clothing and character LoRAs are more compatible. MSG edition is more 水墨畫. I basically turned up the loras in strength until it stopped looking good and then took a step back from that. Stylistic effect is very strong, but it’s still recognizably illustrative. Anime is my personal favorite. It’s more aesthetically 水墨畫 with all that entails but also retains those classic anime style features beneath all the watercolor ink art. You can try style loras on it but to be honest it’s not as versatile as base and might collapse.

Usage Notes:

I recommend you be a bit careful of using Latent upscaler at high denoise. Because the model is a bit heavily opinionated given the style goal and such, at high denoise it tends to shoot strongly to turn it into a real Chinese watercolor painting and thus heavily changes the original image.

With MSG edition, you need to prompt for realism or else the shuimobyhua effect will be too strong. A good style to try is:

(ultra high res,photorealistic,realistic,best quality,photo-realistic),
(high detailed skin,visible pores),(real person,photograph),
(8k, raw photo, best quality, masterpiece)



© 版权声明