Update README.md
Browse files
README.md
CHANGED
|
@@ -38,7 +38,7 @@ MTG card art, Ayula, Queen Among Bears, by Jesper Ejsing, 2019, Green, G, Legend
|
|
| 38 |
MTG card art, Force of Will, by Terese Nielsen, 1996, Blue, U, Instant, uncommon, Alliances, all, Dominaria, Terisiare, Ice Age, expansion, 3UU,
|
| 39 |
|
| 40 |
To briefly explain the entries:
|
| 41 |
-
Every card art is tagged at the start with "MTG card art". Usually you want to use this. It does generalize the image a bit, however. Experiment with using it and not using it. Sometimes, if you are having trouble making something look distinctly "Tarkir" or something, taking off this tag can help de-generalize the art.
|
| 42 |
|
| 43 |
Set type: this is usually "expansion". Other possibilities are "core", "funny", and some other. You can check the Scryfall API documents for more information.
|
| 44 |
|
|
@@ -52,22 +52,126 @@ I like to write my prompts like an art description - you can see in the examples
|
|
| 52 |
|
| 53 |
##Example Images and Prompts
|
| 54 |
|
| 55 |
-
|
|
|
|
|
|
|
| 56 |
|
| 57 |
#Ascended Eldrazi
|
| 58 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
|
| 61 |
#Emrakul, Compleated Doom
|
| 62 |
-
|
|
|
|
|
|
|
| 63 |
|
| 64 |
mtg card art, (emrakul:1.2), (compleated:1.1) doom, (by seb mckinnon:1.1), legendary creature - (phyrexian:1.1) (eldrazi:1.2) (horror:1.1), black, (strixhaven, arcivos:1.2), annihilator, (infect:1.2), 15/15, a (phyrexianized:1.1), compleated Emrakul, attacking (strixhaven school, university campus:1.2), stx, beautiful composition, detailed painting, (sense of scale:1.2), horror, dark, terrifying, eldritch horror, new phyrexia, nph, rise of the eldrazi, roe, extremely detailed, intricate details, masterpiece, best quality, emrakul, the aeons torn, emrakul, the promised end
|
| 65 |
Negative prompt: zendikar, water, ocean, funny, happy, optimistic, bright, tentacles, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, octopus, spikes, urchin, tentacles, arms, hands, legs
|
| 66 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
|
|
|
|
|
|
|
| 68 |
|
| 69 |
## Training and dataset
|
| 70 |
-
Training was done on a dataset consisting of cropped, 512x512 versions of the art for every
|
| 71 |
|
| 72 |
The result is a comprehensive model that has a good understanding of MTG artists, sets, planes, card types, creature types, years, colors, and more. If you had ever wondered what a Merfolk, drawn by Ron Spencer, would have looked like on Tarkir, as part of the Mardu clan, with dash, haste, and trample - this model can deliver what you want.
|
| 73 |
|
|
|
|
| 38 |
MTG card art, Force of Will, by Terese Nielsen, 1996, Blue, U, Instant, uncommon, Alliances, all, Dominaria, Terisiare, Ice Age, expansion, 3UU,
|
| 39 |
|
| 40 |
To briefly explain the entries:
|
| 41 |
+
Every card art is tagged at the start with "MTG card art". Usually you want to use this. It does generalize the image a bit, however. Experiment with using it and not using it. Sometimes, if you are having trouble making something look distinctly "Tarkir" or something, taking off this tag can help de-generalize the art. In a similar fashion, the more general the tag is (ie, rarity, the word "legendary", etc.), the more of a generalizing effect it has on the image. Play around and find out.
|
| 42 |
|
| 43 |
Set type: this is usually "expansion". Other possibilities are "core", "funny", and some other. You can check the Scryfall API documents for more information.
|
| 44 |
|
|
|
|
| 52 |
|
| 53 |
##Example Images and Prompts
|
| 54 |
|
| 55 |
+
This model is trained on so many things, I'm just scratching the surface of figuring out what it can do. I thought it would be helpful to show a gallery of the sort of things I've been able to create with it.
|
| 56 |
+
|
| 57 |
+
Full generation parameters, seeds, etc, should be in the images. All these examples were made with Automatic1111's UI, fantasy-card-diffusion-140000.ckpt, and the "DPM++2S a Karras" sampler. CFG varies - I find around 11 works as a good baseline. Most of these were done with around 40-50 steps - probably overkill.
|
| 58 |
|
| 59 |
#Ascended Eldrazi
|
| 60 |
+
(an Eldrazi that has somehow made his way to Theros, chilled out, and attained godhood)
|
| 61 |
+
|
| 62 |
+
|
| 63 |
+
MTG card art, ascended eldrazi, (by eric deschamps:1.1), (legendary enchantment creature - god:1.2) (eldrazi:1.2), colorless, theros, ths, jou, bng, thb, mythic, indestructible, annihilator, trample, a wise eldrazi titan emerging from the horizon, ascended to godhood, now looking serene, calm, divine, powerful, beautiful composition, emrakul, kozilek, ulamog, (sense of scale:1.2), sense of wonder, overwhelming, extremely detailed, intricate details
|
| 64 |
+
Negative prompt: weak, angry, scary, underwhelming, powerless
|
| 65 |
+
|
| 66 |
+
#Speedy Sliver
|
| 67 |
+
(a Mardu sliver that gives dash, on Tarkir)
|
| 68 |
+
|
| 69 |
+
|
| 70 |
+
MTG card art, speedy sliver, by John avon, Creature - (sliver:1.3), white, black, red, wbr, (Mardu:1.1), Khans of tarkir, ktk, dash, a fast sliver is speeding through the Mardu (steppe:1.1) landscape, beautiful composition
|
| 71 |
+
Negative prompt: human, humanoid, m14
|
| 72 |
|
| 73 |
+
#Taylor Swift, Wandering Bard
|
| 74 |
+
(self explanatory, Taylor Swift, as a bard, on Eldraine. Future Secret Lair?)
|
| 75 |
+
|
| 76 |
+
|
| 77 |
+
mtg card art, (Taylor Swift:1.2), wandering bard, legendary creature - human (bard:1.2), white, red, green, wrg, throne of eldraine, eld, by chris rahn, by volkan baga, by zoltan boros, armored bard taylor swift holding her weapons and instruments, beautiful composition, detailed, realistic fantasy painting, masterpiece, best quality,
|
| 78 |
+
Negative prompt: guitar, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry
|
| 79 |
|
| 80 |
#Emrakul, Compleated Doom
|
| 81 |
+
(The Phyrexians have sprung Emrakul from Innistrad's moon, compleated her, and are now attacking Strixhaven. It's a bad day to go to school.)
|
| 82 |
+
|
| 83 |
+
|
| 84 |
|
| 85 |
mtg card art, (emrakul:1.2), (compleated:1.1) doom, (by seb mckinnon:1.1), legendary creature - (phyrexian:1.1) (eldrazi:1.2) (horror:1.1), black, (strixhaven, arcivos:1.2), annihilator, (infect:1.2), 15/15, a (phyrexianized:1.1), compleated Emrakul, attacking (strixhaven school, university campus:1.2), stx, beautiful composition, detailed painting, (sense of scale:1.2), horror, dark, terrifying, eldritch horror, new phyrexia, nph, rise of the eldrazi, roe, extremely detailed, intricate details, masterpiece, best quality, emrakul, the aeons torn, emrakul, the promised end
|
| 86 |
Negative prompt: zendikar, water, ocean, funny, happy, optimistic, bright, tentacles, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, octopus, spikes, urchin, tentacles, arms, hands, legs
|
| 87 |
|
| 88 |
+
#Ayula, Ascended Bear
|
| 89 |
+
(Ayula, Queen Among Bears is now a Planeswalker, and has taken up residence in Kaladesh)
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
mtg card art, ayula, ascended (bear,:1.1) (by jesper ejsing,:1.1) green, g, legendary planeswalker - (bear:1.1), kaladesh, aether revolt, kld, aer, mythic, beautiful composition, a powerful bear planeswalker riding in a kaladesh (vehicle:1.1), looking very serious, intricate details, ayula, queen among bears, mh1, 2/2, 1g, masterpiece, best quality
|
| 93 |
+
Negative prompt: silly, human, humanoid, breasts, anthropomorphic, bipedal, funny, lowres, text, error, cropped, worst quality, low quality, normal quality, jpeg artifacts, watermark, blurry
|
| 94 |
+
|
| 95 |
+
#Neltharion, Deathwing
|
| 96 |
+
(My attempt at imagining Deathwing as a classic Elder Dragon Legend, with the WoW: Cataclysm Cinematic scene)
|
| 97 |
+
|
| 98 |
+
mtg card art, neltharion, (deathwing:1.2), (by edward beard, jr:1.1), 1994, legendary creature - (elder dragon:1.1), black, red, br, legends, leg, flying, trample, (world of warcraft cataclysm:1.2), large Firey flaming black dragon perched on stormwind castle rampart, roaring, breathing fire, flames, destruction, beautiful composition, extremely detailed, intricate details, masterpiece, best quality, terrifying, epic, cinematic
|
| 99 |
+
Negative prompt: lowres, text, error, cropped, worst quality, low quality, normal quality, jpeg artifacts, watermark, blurry, human, humanoid, deformed, mutant, (ugly:1.3)
|
| 100 |
+
|
| 101 |
+
#Harambe, Simian Champion of Tarkir
|
| 102 |
+
(Harambe did not die, his planeswalker spark ignited.)
|
| 103 |
+
|
| 104 |
+
|
| 105 |
+
|
| 106 |
+
(harambe:1.1), simian champion of tarkir, by magali villeneuve, legendary planeswalker - ape (monk:1.2), white, blue, red, wur, (jeskai:1.2), khans of tarkir, ktk, planeswalker harambe training with the jeskai, in a (monastery:1.2), in the mountains, wearing robes, martial arts, beautiful composition, extremely detailed, intricate details, masterpiece, best quality,
|
| 107 |
+
Negative prompt: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry
|
| 108 |
+
|
| 109 |
+
#Gabe Newell, Techno-Wizard
|
| 110 |
+
(Apologies to Gabe for the prompt - I wanted to make him look kind of like he does today, and the model kept trying to make him look like he did years ago.)
|
| 111 |
+
|
| 112 |
+
|
| 113 |
+
|
| 114 |
+
mtg card art, (gabe newell:1.3), techno-wizard, by zezhou chen, legendary creature - human wizard, blue, red, ur, izzet, ravnica, beautiful composition, (grey beard:1.1), (gray hair:1.1), elderly izzet techno wizard gabe newell is casting a spell, powerful, intelligent, epic composition, cinematic, dramatic, masterpiece, best quality, extremely detailed, intricate details
|
| 115 |
+
Negative prompt: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, young, silly, goofy, funny
|
| 116 |
+
|
| 117 |
+
#Luna, Blind Lunar Goddess of Innistrad's Moon
|
| 118 |
+
(Or maybe just Emrakul in disguise?)
|
| 119 |
+
|
| 120 |
+
|
| 121 |
+
|
| 122 |
+
mtg card art, luna, blind lunar goddess of innistrad's moon, legendary enchantment creature - (god:1.1), by howard lyon, (by chris rahn:1.1), (by seb mckinnon:1.1), (by terese nielsen:0.8), (by rebecca guay:0.8), (by richard kane ferguson:1.1), (innistrad:1.3), dark ascension, shadows over innistrad, inn, dka, soi, white, blue, black, wub, mythic, (blindfolded cute young woman:1.2) as smug (moon goddess:1.1), sitting on throne, dark lighting, full moon night, long white hair, pale skin, (silver blindfold:1.1), opalescent robes, ethereal, celestial, mysterious, beautiful composition
|
| 123 |
+
Negative prompt: orange
|
| 124 |
+
|
| 125 |
+
#Goblin Flamethrower
|
| 126 |
+
|
| 127 |
+
|
| 128 |
+
|
| 129 |
+
(the model can generate instants and sorceries, too)
|
| 130 |
+
mtg card art, (goblin flamethrower:1.1), red, r, instant, sorcery, onslaught, legions, scourge, ons, lgn, scg, a crazed, intense, happy goblin is shooting fire from a flamethrower, dangerous, reckless, beautiful composition
|
| 131 |
+
Negative prompt: (ugly:1.5), lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry
|
| 132 |
+
|
| 133 |
+
#Mox Topaz, Mirage
|
| 134 |
+
(If there had been a Mox Topaz in the Mirage block, drawing inspiration from Volkan Baga's Vintage Masters mox art)
|
| 135 |
+
|
| 136 |
+
|
| 137 |
+
|
| 138 |
+
Mtg card art, two african hands cupped together holding a (mox topaz:1.1) on a gold chain, in the middle of the palm, in front of the (African savannah:1.1), by Terese Nielsen, (by Volkan baga:1.1), by Dan Frazier, artifact, beautiful composition, jamuraa, mirage, mir, vma
|
| 139 |
+
Negative prompt: deformed, bad anatomy
|
| 140 |
+
Steps: 50, Sampler: DPM++ 2S a Karras, CFG
|
| 141 |
+
|
| 142 |
+
#Mox Topaz, Alpha
|
| 143 |
+
(similarly, if there had been a sixth color of Magic, Orange, way back in Alpha)
|
| 144 |
+
|
| 145 |
+
|
| 146 |
+
|
| 147 |
+
(mox topaz:1.1) ( by dan frazier:1.2), artifact, rare, (limited edition alpha, lea:1.1), (1993,:1.1) a mox topaz on a chain
|
| 148 |
+
Negative prompt: lowres, cropped, worst quality, low quality, normal quality, jpeg artifacts, watermark, blurry
|
| 149 |
+
|
| 150 |
+
#Island (Phyrexian Toronto)
|
| 151 |
+
(the Phyrexians have invaded and compleated Toronto)
|
| 152 |
+
|
| 153 |
+
|
| 154 |
+
|
| 155 |
+
mtg card art, (toronto:1.2), (basic land - island:1.1), new phyrexia, nph, by adam paquette, (toronto skyline:1.2), (phyrexian:1.1), dark, horror, cn tower, rogers centre, extremely detailed, intricate details, masterpiece, best quality
|
| 156 |
+
|
| 157 |
+
#Ariel, the Little Mermaid
|
| 158 |
+
(Give it time and I'm sure there will be a secret lair.)
|
| 159 |
+
|
| 160 |
+
|
| 161 |
+
|
| 162 |
+
mtg card art, (ariel, the little mermaid:1.2), legendary creature - (merfolk:1.1), blue, white, red, uwr, (theros:1.1), by Greg Staples, beautiful composition, ariel sitting on a rock with waves, theros temple in background, masterpiece, best quality,
|
| 163 |
+
Negative prompt: green skin, blue skin, red tail, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry
|
| 164 |
+
|
| 165 |
+
#Batman, the Dark Knight
|
| 166 |
+
(Likewise, the Secret Lair is only a matter of time.)
|
| 167 |
+
|
| 168 |
+
|
| 169 |
|
| 170 |
+
mtg card art, batman, the dark knight, by justine cruz, by zoltan boros, legendary creature - human ninja, white, blue, black, (ub:1.1), (dimir,:1.1), (ravnica:1.1), (kamigawa:0.9), neon dynasty, neo, innistrad, investigate, ninjutsu, (at night:1.3), on roof, dark lighting, masterpiece, best quality,
|
| 171 |
+
Negative prompt: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry
|
| 172 |
|
| 173 |
## Training and dataset
|
| 174 |
+
Training was done on a dataset consisting of cropped, 512x512 versions of the art for every MtG card (about 35,000 images), each of which was tagged using a custom python script, from data pulled from Scryfall. Training was done with the Dreambooth extension for Automatic1111's wonderful UI, to 140,000 steps, over the course of a couple days, on my 4090. I changed settings several times as I went, generally increasing batch size and lowering learning rate. At the moment, I am at batch size 10, gradient accumulation 5, and learning rate 4e-7, and that seems to be working well.
|
| 175 |
|
| 176 |
The result is a comprehensive model that has a good understanding of MTG artists, sets, planes, card types, creature types, years, colors, and more. If you had ever wondered what a Merfolk, drawn by Ron Spencer, would have looked like on Tarkir, as part of the Mardu clan, with dash, haste, and trample - this model can deliver what you want.
|
| 177 |
|