This is a safetensors version of a model I trained back in November 2022, likely the first publically available general purpose MTG art model. Surprisingly, I'm not sure if a better general MTG art model has actually been released since November 2022.
I trained this in the early days of Stable Diffusion, before Civitai was really a thing, to my knowledge. I posted a lot of information and a lot of good examples on my Huggingface page for it: volrath50/fantasy-card-diffusion · Hugging Face.
The name, fantasy-card-diffusion, was deliberately vague and doesn't use "MTG" or "Magic", because like a week before I posted the model on Huggingface, WotC shut down some guy's custom card creator, and I figured it was better to go super-generic with the name.
Because I trained it in November 2022, it only "knows" MTG after that point. Sadly, perhaps, I don't think I've actually managed to play MTG since about then , so I don't really know MTG after that point, either. :(
As noted, lots of information and examples on Huggingface, but I'll quote some of the key points:
A comprehensive fine-tuned Stable Diffusion model for generating fantasy trading card style art, trained on all currently available Magic: the Gathering card art (~35k unique pieces of art) to 140,000 steps, using Stable Diffusion v1.5 as a base model. Trained on thousands of concepts, using tags from card data. Has a strong understanding of MTG Artists, planes, sets, colors, card types, creature types and much more.
Prompts: For best results, prompt the model with card information, like you were writing out a custom MtG card, with the phrase "MTG card art" and an art description
Example: "MTG card art, Fiery Merfolk, by Chris Rahn, 2021, creature - merfolk wizard, blue, red, ur, izzet, ravnica, gtp, rtr, grn, an izzet league merfolk, swimming in a ravnica river, casting a fire spell, flames, water, contrast, beautiful composition, intricate details"
Using the Model
The model was trained on MtG card information, not art descriptions. This has the effect of preserving most non-MtG learning intact, allowing you to mix MtG card terms with an art description for great customization.
Each card was trained with card information pulled from Scryfall in the following format:
MTG card art, [Card Name], by [Artist], [year], [colors (words)], [colors (letters)], [card type], [rarity], [set name], [set code], [plane], [set type], [watermark], [mana cost], [security stamp], [power/toughness], [keywords], [promo type], [story spotlight]
A few examples of actual card data in this format:
MTG card art, Ayula, Queen Among Bears, by Jesper Ejsing, 2019, Green, G, Legendary Creature - Bear, rare, Modern Horizons, mh1, draft_innovation, 1G, None, 2/2, Fight,
MTG card art, Force of Will, by Terese Nielsen, 1996, Blue, U, Instant, uncommon, Alliances, all, Dominaria, Terisiare, Ice Age, expansion, 3UU,
Description
FAQ
Comments (8)
I remember this model, it was one of the references I had for my models! Any plans for a future new version? :D
I actually trained newer models over a year ago (one on 2.1, and a second go at training on 1.5) but between not being completely happy with the results, kids, work and ADHD, never managed to get one of those released. If my ADHD focuses long enough, I might release those. They aren't strict upgrades over the OG version.
I also yesterday updated and re-generated my training data, and am making some preliminary attempts to train on SDXL (1 year late to the party), but no idea if those will amount to anything or if I'll get busy with something else before getting an XL model complete.
@volrath50 thanks for all the efforts! I get it too, I want to do things, but with everything else, it takes me too long to progress.
Cheers!
I actually trained newer models over a year ago (one on 2.1, and a second go at training on 1.5) but between not being completely happy with the results, kids, work and ADHD, never managed to get one of those released. If my ADHD focuses long enough, I might release those. They aren't strict upgrades over the OG version.
I also yesterday updated and re-generated my training data, and am making some preliminary attempts to train on SDXL (1 year late to the party), but no idea if those will amount to anything or if I'll get busy with something else before getting an XL model complete.
Sweet checkpoint. But I have a question.
Output seems to be cut off on the right a lot. Is that some general mtg card art bias? Or something went wrong with cropping?
Yeah, that's because when I trained this 1.5 years ago, I did a quick resize/crop to 512x512, and cropped a bit of the right off. (There weren't as many tools to easily do cropping, and bucketing wasn't a thing yet.) It was meant to be a test run, but turned out better than expected and I released it.
I fixed it in the unreleased versions I trained, but those had other problems. I still have meant to release them for well over a year, but got busy with life and stuff. Hopefully while my attention is back on my models, I'll manage to release them before my ADHD gets distracted with something else.
@volrath50 I would like to see the updated 1.5 version. These are the hidden gems of Stable Diffusion. You managed to do all this despite ADHD.. well done my dude!
Brings back great memories of discovering Stable Diffusion. Would love an SDXL refresh if you ever feel inspired
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.