All credit goes to @Minthybasis for his finetune and the extract, and the authors of the original models.

I merged the latter into some models and found that it learns more tags, improves the image & contrast.

More tags from danbooru like artists, characters, styles, and general concepts are added.

Resume of the versions:

Azoth is my latest merge and will handle multiple styles/lora combinations. Arael is the same but with flatter colors.

Chacol is a mix that has ebara and various loras. This one tends to make more backgrounds than the others.

Confetti contains regular pony, autismmix, js2prony and confettitune.

Tcs has more shading. Tko contains some loras, can be unstable.

Adding the lyco will work using this node, these settings.

Using a normal "Load Lora" node will not have any benefit, since it's placing the data over the entire model.

By writing to the output blocks using the lycoris preset, the data contained within the extract does not overwrite Pony's existing data of anatomy and composition, instead adding it to blocks outside of them. But it does improve the anatomy in my testing.

Here is a study of the artists contained in 4thTail. Some are less pronounced than others. This is a manual study so there may be more. Characters sorted alphabetically.

If a model has special training data for styles, like too many loras, it'll overwrite it (it goes over the style blocks), so i don't recommend it. I did not merge it into Insomnie because it gets rid of the dithering. Same with realism models.