A powerful LoRA trained on modified photos taken by myself of all 78 of a Rider-Waite tarot deck images originally from 1909, based on Golden Dawn symbolism.
This LoRA was trained at 768x1344 but works great at most sizes, I recommend: 1024x1400.
Text will often double or miss letters, but text will almost always be added in the correct locations on the cards. Often very simple to fix in an external photo editor. On the bright side, it forces you to fix and alter the image enough to perhaps claim a level of ownership on the created images.
Will update if I can get more reliable text with training.
Works best with simple prompts under five words. Use lesser strength prompts after the words you want to appear to add more control to your card.
"trtcrd artstyle prompt here"
"trtcrd artstyle queen of derp"
"trtcrd artstyle king of emeralds [elon musk]"
☕ Like what I do? ☕
☕ Buy me a coffee or two! ☕