santa hat
deerdeer nosedeer glow
Sign In

Headbone Interactive (Elroy) Style

13
69
550
6
Verified:
SafeTensor
Type
LoRA
Stats
69
550
Reviews
Published
Jul 16, 2024
Base Model
Pony
Training
Steps: 700
Epochs: 10
Usage Tips
Clip Skip: 1
Strength: 1
Trigger Words
headbone_style
source_cartoon, pixel art, flat color, orange skin,dot eyes, no sclera
dithering, photo background, realistic background, monochrome background
Hash
AutoV2
DB1939432A
Created on Civitai

Headbone Interactive was a short lived (founded 1994, releases 1995-1997) Seattle based video game studio that made Shockwave based educational adventure games for CD release. Their games featured an unusual art style that combined monochromatic backgrounds created from composited photographs (a technique deemed “photocollage” by the credits), that house flat colored cartoon characters (and interactive elements) on top. This LoRA was trained on 175 hand tagged images from the games Elroy Goes Bugzerk, Elroy Hits the Pavement, Elroy’s Costume Closet, and Iz and Auggie Escape From Dimension Q (Headbone’s Gigglebone Gang series features a different artstyle by different artists and was not used for training). I am aware of the pilot for Elroy: The Animated Series, but the only surviving copy has heavy compression artifacts unsuitable for a pixelart LoRA (also Elroy with sclera is kinda creepy). I didn’t have the downloadable costumes for Costume Closet (and don’t know if they even exist anymore, please tell me if they do) and thus they aren’t in the training.
Trigger Word: headbone_style

Character prompt: source_cartoon, pixel art, flat color, orange skin,dot eyes, no sclera (swap last two with “black eyes” for Dimension Q style. While orange skin is the default, some tertiary characters with “pale skin” and “dark skin” were trained, as were a few “gray skin” beings from Dimension Q)

Background prompt: dithering, photo background, realistic background, monochrome background, indoors/outdoors (you NEED to specify background elements like a type of place and some objects, even just "Simple Background, white background" for a character alone, or you get a corrupted background. Removing "dithering" from the prompt may make certain background types less corrupted. Outdoors and indoors were fully tagged and seem to make backgrounds much more stable. "Sky" can also help for outdoors)
Negative: Ears, realistic, gray skin, robot, off center, in background, distant

Characters:
Elroy series:
Elroy: 1boy, elroy_headbone, brown hair, short hair, Red hat, baseball cap
Bugzerk outfit: yellow shirt, green overall shorts, red shoes, sneakers
Hits the Pavement outfit: backwards hat, orange shirt, striped shirt, dark green shorts, red shoes, sneakers
Sydney DelGado: 1girl, Syd_headbone, flat chest, black hair, bowl cut, hairband, purple hairband
Hits the Pavement outfit: pendant, pink dress, purple tights, brown shoes
Costume Closet intro outfit: magenta tanktop, pink belt, blue jeans, blue shoes,
(As so many images of Elroy and Syd are from Costume Closet, try using “wig” as a negative)
Blue: Blue_Headbone, 1other, dog, blue fur, bloodhound, collar
Costume Closet:
Cecil: 1boy, Cecil_headbone, pale skin, bald, black beret, mustache, monocle, white shirt, long sleeves, sleeve cuffs, teal vest, blue ascot, wristwatch, black pants
Escape from Dimension Q:
Izabel LeGrand: Iz_Headbone, 1girl, flat chest, light blue hair, bob cut, purple shirt, long sleeves under short sleeves, blue sleeves, maroon pants, black boots (may need "Gray Skin" in negatives)
Augustus T. Robot (Auggie): Auggie_Headbone, robot, wheel, yellow eyes, floating head

Try specifying colors for foreground objects. The nameless mobsters of hit the pavement gave ample training data for “stubble”. Most characters in training data had four fingers (3+thumb), so that’s not a defect, it’s a feature. Set character LoRAs to 0.80 strength or less.


Checkpoint: AutismMix produces really good character results or really terrible ones (very rarely mediocre) while struggling with backgrounds. Pony 🦄FaeTality seems to be most consistent with not totally wrecking the backgrounds and produces characters that look good, though aren't always 100% in style (limbs are thicker than they should be, hands are overly detailed).

PerfectPony mutes the style and can't do the backgrounds at all. AAAAutism does poor all around (if not totally corrupted backgrounds). Pottingmix PDXL just doesn't get the style at all. Base Pony is AutismMix but worse. ZavyfantasiaXL does OK but not great backgrounds and doesn't understand the character style. Bluvoll's 3pxDiffusion is mediocre. Cashmoney (Anime) does above average backgrounds and below average characters, might be worth more experimentation.