Janky experimental LoRA trained on various "horny warning sign" type images.

To trigger use one of the following styles:

black background, black lineart, white body - 17 images in dataset

yellow background, yellow lineart, black body - 8 images in dataset

yellow background, black lineart, yellow body - only 2 images lmao

other color combinations can be used, but may not always work.

humanoid can be used to trigger "warning sign person"

! IMPORTANT !

3d in negative helps tremendously to keep the style "flat".

Lowering weight improves flexibility.

Don't use too many "score" or "quality" keywords in the prompt. It might break simplistic style and add unnecessary shading or detail.

Use flat colors, lineless, simple background etc. as necessary to help with the style.

Version 1.1 changes:

• cleaned up dataset a bit more;

• trained for longer.

As a result, LoRA seems to be better at generating less details and keeping the image "flat". Still janky, lol.

This version is available in 2 variants: 1.1a and 1.1b

1.1a generates less detail than 1.0

1.1b has been trained even longer and tends to remove even more detail

Check this post for comparsion grids