MadMix - beyond the good taste

Details

Download Files

About this version

The suggested serie of embeddings is a good start to adapt the style. Tested on IL and NAI. Then you can add tags as "flat colors", "source comic" (be aware, add "multiple views" and/or "multiple frames" in the negative in that case), adjust the negatives (such as "volumetric shadowing") etc. to have your toon graphic.

OR you can do the reverse, asking for volumetric shadowing and putting the 2D embedding in the negative.

ALWAYS control what the embeddings add to the prompt, or else younwill get a bad case of blurry eyes and artifacts. But really, that serie is THE most reliable I found. DO NOT use the Smooth serie, those are worse than going without embeddings; like, really really bad, at least in my experience.

Model description

Not furry: the examples are just a by-product of my personal taste

As usual for my mixes, the influence on the shapes and colors is subtle: crank up the strenght if necessary.

An overachiever, this LoRA. It leans heavily on the model, but influence everything in the generation, usually for the better (relevant exception, it doesn't like much the diagonal axis and characters will try a bit to follow vertical and horizontal lines).

Were it shines bright, from the first tests, it's in distinguishing 2D styles and 3D styles, see the above examples with the catgirl and note changes in the prompts (without this LoRA, the 3D wasn't consistent and the 2D... Wasn't 2D). The output can be really eterogeneous, depending on the prompt (or embeddings) and the model (more anime model will take less from toons, and vice-versa).

Trained on comic toons from two authors, an old hcg (sex friend, if you are interested), some rendering from two sources and, finally, some furry pic to be ready for everything.

Yes, I'm mad. No, I regret nothing.

Images made by this model

Sort by

No Images Found.

Model Type	LORA
Base Model	Illustrious
Published	7/25/2025