YukichanMix

About this checkpoint

This checkpoint is my own variant built starting from WaiSnackMix by @MGHerder.

I want to sincerely thank him, because he helped me understand the merge process much better and also gave me the chance to use his work as a starting point. Without his help and without the work done on WaiSnackMix, this version would not exist in its current form.

Starting from that merge, I created my own interpretation of the model by baking into it a combination of LoRAs, pushing the output toward a direction closer to my personal taste and the kind of images I wanted to achieve.

A significant part of the credit therefore goes to @MGHerder, both for creating the original merge and for helping me understand and use it better.

---

What to expect from this version

Compared to the original WaiSnackMix, this variant has its own identity due to the integrated LoRAs and the merge direction I chose.

The goal was to keep good stability and solid output quality, but with a different stylistic influence compared to the original version. Because of that, the model may behave differently in areas such as:

• style intensity

• rendering of certain details

• behavior on clothes, materials, and textures

• impact of certain prompt keywords

This version should therefore be seen as a personal reinterpretation of the original merge, not as a simple reupload.

---

Prompting and keywords

The checkpoint still responds well to both booru tags and more descriptive prompts.

Since this variant is derived from WaiSnackMix, I am also including a few additional tag suggestions directly inspired by the original checkpoint description from @MGHerder, as they can be useful as a starting point:

For more detail, polish, and shading
masterpiece, amazing quality, best quality, ultra-detailed

For a more realistic direction
masterpiece, realistic, amazing quality, best quality, ultra-detailed

For a more toon-like look
toon style

Other lighter stylistic modifiers
anime style, retro style, kawaii style

Tags that can shift the style more noticeably
vibrant hues, dynamic pose, blue palette, realistic, abstract background

They are not required, but they can help you better guide the output.

---

Settings

I am still testing this variant, so for now I do not want to lock in any definitive settings.

You can either try the settings you usually use, or test the ones I used below and adjust from there depending on the kind of image you want.

One setup that worked well in my tests was:

Euler Ancestral + Simple

• CFG: 5

• Steps: 30

I also got very good results with:

DPM++ 3M SDE + AYS SDXL

• at least 50 steps

• using Dynamic Thresholding and Skimmed CFG

With this second setup, if the prompt is written well, it is possible to use higher CFG values, in some cases even going up to 10.

It also works very well with Euler Ancestral, and as schedulers I recommend trying both Simple and Karras, depending on the type of output you want.

I encourage you to experiment a bit, because the model can react quite differently depending on the prompt and the scene.

A proper upscale is strongly recommended, and for maximum quality, a very useful tool can be ADetailer, especially for improving the eyes.

---

Recommended image size

I also recommend generating images at a relatively high base resolution, as this often allows the model to produce a greater amount of detail right from the start. In many cases, this can result in images that already look strong without requiring upscaling, while also behaving better if you do decide to upscale them afterward.

In my case, my standard base generation size is 1024×1536.

---

Final thanks

Another thank you to @MGHerder for the help, the explanations, and for creating the merge this version was built from.

My contribution in this version lies in the merge direction and in the combination of the integrated LoRAs, but the foundation of the work comes from his checkpoint.

If you like this variant, I also recommend checking out the original WaiSnackMix and supporting its creator.

---

Checkpoint License & Terms of Use

Use of this checkpoint must remain in compliance with the Illustrious model license and usage guidelines.

User Responsibility:

You are fully responsible for any images or other outputs generated with this checkpoint, including how it is used, prompted, shared, or distributed.

Restrictions:

No use for unlawful, harmful, or abusive purposes.

No use in violation of the Illustrious license, rules, or community guidelines.

Any redistribution, merge, or modified version must remain compliant with the original applicable license terms.

Failure to comply with these terms voids any right to use this checkpoint.