This Lora creates playing card symbols
Prompting:
"a set of poker symbols, a spade, a heart, a club, a diamond,"
simple/NOTHING/detailed/very detailed/ultra detailed
white background/black background/negative space
"cat theme" or "celtic theme" or "gothic theme" or "floral pattern" or "scribbles pattern" (the difference in the training data between pattern and theme is that you can see the elements of the theme (you'll see a cat or a tree) whereas a pattern is just the vague shape of the tiny scribbles)
"3d" for a bit of 3d, "outline" for outline only. "reflection" for glossy, "gradient" for a color gradient
Notes:
trained on SD base 1.5
works on realistic models but doesn't seem to work on anime models (raising the lora weight gives something reasonable but you lose the style of the underlying checkpoint)
there is no italian/spanish style symbol (cups etc...) because I have never played with those cards
it is customary to have simple symbols for cards and more complicated one for aces (reminder)
Description
FAQ
Comments (4)
LOVE this idea!!! Though, Could you now do a LoRA for the face cards? King, Queen, Jack, and Joker. Please.
see suggested resources (hint: this doesn't work properly because centro symmetry is a concept stable diffusion doesn't understand correctly)
@raphb Oh darn. Maybe train them tilted slightly? Still probably wouldn't work.
@CupcakeDreamer I think it would require something like the 'seamless tileable' hack. The author (maybe Thomas Moore) told the convolution layers (they analyse the image by making a weighted sum of nearby pixel) to wrap around when they reach the border of the image instead of repeating the border pixel. See line 24 https://github.com/TomMoore515/material_stable_diffusion/blob/main/predict.py .










