Great work! love your checkpoint.

can you please upload a bakedVAE version?

Thank you in advance.

真他媽漂亮

大顆壽司的創作者哥哥（姐姐？）偶關注泥很久了，沒想到泥偷偷幫我在2.5DSET 的留言點讚？ XD 希望哥哥（姐姐？）也可以合一個真人版模型？ 你們兩個合起來一定會非常的漂亮 😍😍

I experimented a lot with this model and found the ideal settings to get images like the examples provided by the creator.

VAE : vae-ft-mse-840000-ema

Clip skip: 2

CFG: 10

Latent upscale: 0.5

Upscale by 2x (This is vital)

Image must be at least 512x512 to start.

Without upscaling by 2x the images colors will appear washed out. I normally render images without upscaling 2x so it took awhile to find this.

This is also for the colorful and darker models not the 2.5d one so I'm not sure if these settings would work on 2.25d model.

Gorgeous model.

大佬，请问一下为什么我拷贝下来，鼻子和嘴巴很抽象啊，换着试了很多次了，基本都是鼻子和嘴巴的问题。

is it ok to use this model in commercial?
Are there any ristrictions?

画躺着和坐着时有遮挡关系的腿时有点灾难

希望出个小版本改进一下，真的太容易崩了

thanks for sharing this checkpoint your two sushimix are my favorite ones, the results are amazing!

希望能有自带VAE的版本

好像不能100%还原，作者大大能说下vae是哪个吗

没有ckpt吗？下完之后不会用啊，safetensors文件不会用啊。

感觉鼻子有点不对劲，默认好像只有鼻头，没有鼻梁。鼻梁感觉被弱化了，能优化吗。还是我设置问题。

Prompt Guide for Dark Sushi Mix

https://www.youtube.com/watch?v=hrrvZBNn4U8

noiseoffset?废玩烂梗

想知道这种训练的原始数据是从何而来的

we really need this for SDXL :)

请问作者可以在主页增加一个社交网站吗，想联系您合作

Images are obtained in gray tons, help

大颗寿司我的挚爱啊！！！！！大大还有群吗！！！我想进去一起交流！！！！

大神能支持sdxl turbo么，速度更快。感谢感谢

這個不讚不行

nice model! but please fix the hands if possible.

whoa... this checkpoint definitely has a type.

eye spacing, eye size, nose size, complexion shading hints, expressions, body shape size and dimensions across-the-board.

it's quite specific; it's like nvinkpunk, where you type in a prompt of two nonsense words, and get a close-up portrait or a futurist cityscape. Sometimes gorgeous ones! But that's not the point. ❣️

The best anime model I've seen so far. It produces amazing results! <3

超级无敌喜欢你的模型！！！

题外话：王国之泪不看攻略的话一个月打不完的吧，好好享受的玩儿吧~

nice model!

How are the hands fixed?

Nice model!

I'm going to be honest, I'm pretty new to AI generation but after playing with many models, I gotta say this is definitely one of my favorites...

It just has so much versatility in anime styles, and plays nice with lots of different loras. Also just has a beautiful style of its own.

Like I said, definitely one of my favorites besides Pony and AnythingXL when it comes to anime stuffs

I already left a comment on here reviewing how good this checkpoint is, however it seems to fall a little bit short when it comes to generating hands

so for anyone who might be struggling, these embeddings work pretty good =w=

https://civitai.com/models/16993?modelVersionId=20068

https://civitai.com/models/56519/negativehand-negative-embedding

Sometimes both at the same time, sometimes one or the other

manually fixing the hands and putting them back in through img2img also works very well

I'm trying to reproduce the texture of the picture like the barefoot girl in the preview image, but the texture of the output image is different. The model used is the same, of course, and the sampler and prompts are almost identical. I really like the texture of the picture in the preview image, so I need someone to tell me how I can reproduce it.

Excellent Anime Checkpoint, one of the best I've come across.

想请问一下作者，在使用大颗寿司模型时候，人物手上总是会出现水彩笔，这个有办法解决吗？我在正反向关键词里面都添加了相应的去除，但是在生成的图片上还是出现水彩笔

I love the art style of this model, but everything I generate with it looks horribly burned unless I use the grapefruit vae. I'm sure there's something wrong on my side, I just have to experiment more.

Sick mix, Lighting is so good, flexible, clothing is great !!!

dfekg

It says "use highres fix to improve quality." I'm new. What is this highres fix everyone keeps talking about? Thanks :)

vary good

could you make an update on this checkpoint? i've been using it a lot on @Daily_waifus_club, i really love it

This model is great, and it matches the style of lora I trained myself. I like it very much. The only drawback is that the fingers are very easy to break. I trained several loras but it didn't work.

great work