Information

Models mainly for anime. Compatible with pony loras.

Knows more tags from danbooru such as; characters, styles, concepts. Maybe some artists.

The additional knowledge comes from this lora. I merged only the output blocks with the lycoris, so the anatomy from the base models is not compromised.

Example images have no negative prompt.

Prompts

- You don't need the score tags in positive.

- More responsive to style tags, examples;

graphite pencil, 2.5d, bold line, smooth shading, flat colors, minimalistic, cgi, digital painting, ink style, oil style, pastel style

- Negative prompt can be left blank. If you want a negative, don't use long ones.

> Something like "abstract,3d" or "text,watermark" is good. "photorealistic,photo \(medium\)" for cartoon, "text,manga,sfx" for less speech bubbles. Also a few keywords for unwanted objects.

Tips

- If a character isn't picked up by it's name, you can use their physical description and copyright.

Example: neptune_\(neptunia\), neptune \(series\), d-pad hair ornament, hooded track jacket

- Perturbed attention guidance improves details, scale near 1.

Nodes used with ksampler on example images:

RecenterXL + PerturbedAttentionGuidance