PopSkull

Why haven't i updated the model or gone for a newer flexable model? cost. if you want me to train more i need more buzz. not asking for money. but if you have buzz then it makes it easier for me to train.

Recommended settings

1. Steps: 20,

2. Sampler: DPM++ 2M Karras,

3. Guidance Scale: 5.0,

4. Size:512x768 then upscale

5. Clipskip: I can’t decide

6. 1: more saturated, follows the prompts a bit better

7. Tends to look a bit too clean sometimes

8. 2. More desaturated, follows danbooru tagging a bit better, has an older comic feel to it, also a few more styles come through better in the mix with this.

1.7 new artist styles added.

Hasui Kawase

Irma Stern

Josh Adamski

Marianne North

Mati Klarwein

Maud Lewis

Marianne von Werefkin

Max Beckmann

Michael Parkes

Mikalojus Konstantinas Ciurlionis

Nicolas Delort

Remedios Varo

Stanley Donwood

Tomasz Alen Kopera

Tatsuro Kiuchi

Victor Brauner

Victor Moscoso

Virgil Finlay

Philip Guston

Margaret Macdonald Mackintosh

Noriyoshi Ohrai

Peter Max

(simple 1 click vote to Tell me which checkpoint suits your needs the best so i know which one to merge back in on further expansions)here

if the survey doesnt work let me know via chat, ill try to figure out a way to show the results here. im not trying to market anything, just thought it would be nice to know what people thought.

1.6 - info to come, seems clip skip 2 is working better with this update, trained in some midjourney painter and cartoonist sample images, as well as some some artist specific loras that I trained. will try to document more later.

If you like the model and want to give a cup of coffee, you can do that here.

1.1- gave it a realistic burnout by laying in epicCartoon underneath like a low pile carpet, no offensive to it, but used to get rid of a lot of the realism left over from my trained model that I lora'd on top of. epicCartoon gets super big props for this as it should have been the model that this was built off of but I didn't know it existed. It doesn't change most of the model but helps with further training.

0.95

cleaned a lot, which unfortunately removed most of the trained data but hopefully will be worth it when retraining on this version, will post again hopefully its the v1.0

you may see "Chode***" in the prompts of the some of the sample images because that was the original name of what the PopSkull model was going to be called CHODE an acronym for Curated Hauntings of Divergent Expressions. ;)

****something something letters symbols things

POPSKULL 0.5 BETA

0.85 may be a bit too heavy on the grime. I feel like I maybe have lost some of the fun. let me know.

SURREAL

DARK FANTASY

LOWBROW

LOWBROW ART

POSSIBLE COLLAGE

ARTISTS

HORROR

POSTMODERNISM

POP SURREALISM

POPART

KITSCH

**Antiquated, only applies to the first of the betas***

I’ve added a helpful negative embedding for this model as a zip file. It’s listed as a checkpoint because there wasn’t another option. It’s only a 5 vector TI so it shouldn’t weigh down the negative too much.

try including words like lowbrow, pop surrealism, and artists that you like