As for "guardian *****", yes, my networks curse too. I swear that they have come up with hideously offensive names like "*****-fire elemental" (just picture that in your mind), and I've avoided mentioning them thus far because I didn't want to detract from the successes
So far, my girlfriend's favorite card name that my RNN has generated is "Morti****er."
6B
death strength
instant
target player discards a card.
= put a 2/2 black zombie creature token onto the battlefield.
= put a +1/+1 counter on target creature.
= destroy target nonblack creature. it can't be regenerated. you gain life equal to its toughness.
At temperature 0.5, that's very good, honestly, and that brings up an interesting point. At lower temperatures, I have the feeling that the network is more reliably producing cards with abilities and effects that are on point in terms of color. Of course, that's just from what I see, I'm sure we can measure the in some way to confirm that. Still, that's very promising. I'll have to write a script later for mid-sequence priming based on the input format. I'm very curious to see what multicolored creatures look like. Since they're rarer occurrences, it might just give a hazy mix of multicolored text that may or may not be relevant to the creatures colors. For example, if I do primetext "|crafty merfolk||creature||merfolk rogue|&^^/&^^|{^UG}|" and I sample at temperature 0.5, I can get cards like
Crafty Merfolk UG
Crafty Merfolk can't be blocked.
When Crafty Merfolk enters the battlefield, sacrifice it unless you sacrifice a black or red permanent.
2/2
Emphasis mine. Notice that you can get cards that do multicolored things but the colors involved aren't necessarily the colors of the creature.
On the other hand, it's difficult to track what the network is thinking because if you prime with "|mistfire blade||creature||human warrior spellshaper|&^^^/&^|{^^^RU}" you can also get cards like
Mistfire Blade 3UR
Creature - Human Warrior 2UU, T, discard a card: draw a card.
3/1
Which is fine for a red creature. However, if I prime it with the colors in the mana cost reversed (blue after red), with the same seed, temperature, and everything, I get...
Mistfire Blade 3RU
Creature - Human Warrior 3, T: Target creature gains flying until end of turn.
3/1
We get an effect that is more blue than red, which implies that the network thinks that the last mana symbol, which is closest to the body text, is most important to determining what should be written in the card. Since the majority of creatures are monocolored, that rule makes perfect sense. The network usually converges on the least complicated rule for predicting the text, so if it can get away with counting the number of mana symbols (how big or splashy the card needs to be) and the type of the last mana symbol (what color the card needs to be), then it's going to do that.
The trick then, I feel, is more robust input generation for the training process. We have to force it into situations where it gets the card wrong unless it takes all of the colors into consideration. I think we can do that.
For those of you who want a comparison in recent literature, there's Zaremba and Sutskever 2015. In that paper, the authors teach a 2 layer LSTM network with 400 neurons per layer that reads in characters one at a time (in short, much like ours) to simulate the execution of Python code. The authors found that the secret to getting it to understand Python code is to feed the network simple examples of generated programs and then build up to more complicated generated ones, a strategy called curriculum learning.
Following with that reasoning, I previously implemented a generator earlier for simple creatures to help the network grasp relationships between mana costs, abilities, and power and toughness, and I was reasonably successful with that. I think the next step is to extend that type of approach to more card types, multicolored cards, etc.
With a good encoding (for which we have several good candidates), good card generation (for which we have some work done already), and a good training curriculum (for which we have support in the literature), we should be able to up the quality of the average card produced by the network considerably. From there, careful application of script-driven priming should get us the rest of the way to having full-fledged sets. That's the trajectory I have in mind.
I'm going to have so much fun drafting these cards.
1R
scroll of giant
creature - goblin warrior
inspired ~ whenever @ becomes untapped, tap all creatures you control.
2/2
This guy is actually super neat, although would work better as a bigger creature that taps all other creatures. I like that within all the random junk there are tiny gems of elegance like this one.
The NN would probably have an easier time learning X cards if you used a specific token that appeared only for the X value, wouldn't it?
The problem is related to coreference resolution. For example, say you have the following English sentence:
"John left the house. He went straight to work."
"John" and "He" are coreferents, in this case, "John" is the antecedent which defines "He". You can have the first sentence without the second and have everything be syntactically and semantically fine, but the second sentence is missing semantic information without the first (who does "He" refer to?).
There are many examples of coreference in Magic text, here are some examples:
* "Counter target spell. Its controller may draw up to two cards at the beginning of the next turn's upkeep."
* "Kicker3 ... If Æther Figment was kicked, it enters the battlefield with two +1/+1 counters on it."
* "Enchant creature ... Enchanted creature gets +1/+2"
Now, the network has to learn from scratch, with no prior knowledge about what the text means or what relationships exist. Now, it does a decent job of picking up on those kinds of relationships when it has lots of examples, like aura cards or equipment cards. It sometimes falls short on others:
2R
shinen of savagery
creature - elemental warrior
kicker 1G and/or U
when @ enters the battlefield, return a blue or red creature you control to its owner's hand. T: add W to your mana pool.
2/2
GW
wall of falter
creature - troll cleric
whenever @ deals combat damage to a player, put five +1/+1 counters on it unless it was kicked.
2/2
And this same problem is showing up with X costs. These are especially tough situations where you have two coreferents that depend upon each other to be meaningful. Of course, it's fixable with the right training.
The NN would probably have an easier time learning X cards if you used a specific token that appeared only for the X value, wouldn't it?
The problem is related to coreference resolution. For example, say you have the following English sentence:
"John left the house. He went straight to work."
"John" and "He" are coreferents, in this case, "John" is the antecedent which defines "He". You can have the first sentence without the second and have everything be syntactically and semantically fine, but the second sentence is missing semantic information without the first (who does "He" refer to?).
There are many examples of coreference in Magic text, here are some examples:
* "Counter target spell. Its controller may draw up to two cards at the beginning of the next turn's upkeep."
* "Kicker3 ... If Æther Figment was kicked, it enters the battlefield with two +1/+1 counters on it."
* "Enchant creature ... Enchanted creature gets +1/+2"
Now, the network has to learn from scratch, with no prior knowledge about what the text means or what relationships exist. Now, it does a decent job of picking up on those kinds of relationships when it has lots of examples, like aura cards or equipment cards. It sometimes falls short on others:
2R
shinen of savagery
creature - elemental warrior
kicker 1G and/or U
when @ enters the battlefield, return a blue or red creature you control to its owner's hand. T: add W to your mana pool.
2/2
GW
wall of falter
creature - troll cleric
whenever @ deals combat damage to a player, put five +1/+1 counters on it unless it was kicked.
2/2
And this same problem is showing up with X costs. These are especially tough situations where you have two coreferents that depend upon each other to be meaningful. Of course, it's fixable with the right training.
With kicker, that makes some sense, as a simple gatherer search for "kicker" returned only 113 cards that have kicker (this includes multi-kicker). There are 236 cards with X in the mana cost, and many more that have both occurrences in the rules text. I would have expected there to be enough X cards for it to figure it out. In any case, generating large amounts of cards with kicker (and other abilities that have this coreferent problem) seems to be the way to deal with this. I suppose the trick to that is making sure those 'training' cards aren't skewing the power level for cards with that ability.
With kicker, that makes some sense, as a simple gatherer search for "kicker" returned only 113 cards that have kicker (this includes multi-kicker). There are 236 cards with X in the mana cost, and many more that have both occurrences in the rules text. I would have expected there to be enough X cards for it to figure it out. In any case, generating large amounts of cards with kicker (and other abilities that have this coreferent problem) seems to be the way to deal with this. I suppose the trick to that is making sure those 'training' cards aren't skewing the power level for cards with that ability.
My thought exactly. For example, it's fine with generating cards that have kicker and it's fine with generating cards that have "if kicked" clauses, we just need to teach it you can't have one without the other, that there is some kind of mutual dependency between them. Feeding it additional generated cards could solve that problem.
Heck, we can generate those cards with priming, actually. If I prime the card to generate "Kicker <cost>" followed by an "if kicked clause", it can fill in what happens when the conditions are met. That may be one way of training it to recognize that relationship, with a feedback loop at low temperature. I might try that in the near future and see what I get.
The NN would probably have an easier time learning X cards if you used a specific token that appeared only for the X value, wouldn't it?
Indeed, and unless I'm mistaken, I am already doing this, because the text is all lowercase. The only uppercase letters are for mana symbols and occurrences of X, which looks the same inside and outside of mana symbols (though to be consistent with other mana symbols, X in a mana symbol is actually written XX).
One thing I noticed about hardcast_sixdrop's cards is that the network appears to be fumbling toward grasping how Traps work? it only generated two which is hardly a large sample, but both have clauses of the form "If [condition], you may pay [cost]". It's phrasing them as an additional rather than an alternate cost and neither spell does anything if that cost isn't paid, but it's still pretty good considering how few Traps actually exist in the game
(also hello I've been lurking the thread for some time but for whatever reason this is the first time I'm moved to comment)
ETA: also from h_s's cards wow, that network sure has a thing for extremely expensive black sorceries doesn't it?
I still think we might be approaching this issue with "X" with too much outside knowledge. Remember the difficulty it had trying to remember what types of counters to use in different situations? X is a variable that we use it in too many places and it is too difficult for the machine to understand that the meaning of X changes depending on the context in which it is being used. We could instead differentiate the behavior of X into different situational abbreviations. I'm sure you can think of more variables that X can be used, but for my example I'm using 'X', 'XA', and 'XY'.
X = X as a mana cost, and abilities, counters that are directly related from X mana being paid in the casting cost.
Example: Fireball deals X damage, or Creature comes into play with X +1/+1 counters on it.
XA : X as a paid ability, when the casting cost of the card has a set converted mana cost.
Example: Creature costs 1MM, has ability "Pay XA: Sacrifice an artifact, ThisCard deals XA damage to target creature or player."
XY = X is dependent on a secondary value or condition.
Example: Creature enters the battlefield with XY +1/+1 counters on it where XY is the number of cards in your hand.
Warbrank Stalker 4BB
Creature - Vampire Minion T: Consult the sacred scriptures for guidance. If you do, at the beginning of the next end step, put a 2/2 black Angel creature token with flying onto the battlefield.
2/4
#And our prayers are answered at last!
I still think we might be approaching this issue with "X" with too much outside knowledge. Remember the difficulty it had trying to remember what types of counters to use in different situations? X is a variable that we use it in too many places and it is too difficult for the machine to understand that the meaning of X changes depending on the context in which it is being used. We could instead differentiate the behavior of X into different situational abbreviations. I'm sure you can think of more variables that X can be used, but for my example I'm using 'X', 'XA', and 'XY'.
X = X as a mana cost, and abilities, counters that are directly related from X mana being paid in the casting cost.
Example: Fireball deals X damage, or Creature comes into play with X +1/+1 counters on it.
XA : X as a paid ability, when the casting cost of the card has a set converted mana cost.
Example: Creature costs 1MM, has ability "Pay XA: Sacrifice an artifact, ThisCard deals XA damage to target creature or player."
XY = X is dependent on a secondary value or condition.
Example: Creature enters the battlefield with XY +1/+1 counters on it where XY is the number of cards in your hand.
That's a possibility, considering X can be used in multiple contexts. It would definitely lessen the cognitive burden on the network, which makes it more likely to learn the relationship. Still, if we make more progress on the "kicker/kicked" front, that should ameliorate the quality of our results. Or so I hope, I'm really just hypothesizing at this point until I get more data to back it up... and data I shall have, soon.
Warbrank Stalker 4BB
Creature - Vampire Minion T: Consult the sacred scriptures for guidance. If you do, at the beginning of the next end step, put a 2/2 black Angel creature token with flying onto the battlefield.
2/4
#And our prayers are answered at last!
HAHAHAHA
Can this thread just be stickied already
I'm glad you enjoy my sense of humor, lol.
EDIT: I was having some fun just a moment ago...
Loyal Hornemaker 1RR
Creature - Goblin Artificer
Renown 2 (When this creature deals combat damage to a player, if it isn't renowned, put two +1/+1 counters on it and it becomes renowned.)
When Loyal Hornemaker becomes renowned, remove a +1/+1 counter from it.
1/1
#Not quite what I had in mind, but at least it understood that Loyal Hornemaker could have counters to take away.
Hardowed Healer 3UU
Creature - Spirit
Renown 2 (When this creature deals combat damage to a player, if it isn't renowned, put two +1/+1 counters on it and it becomes renowned.)
When Hardowed Healer becomes renowned, remove all +1/+1 counters from target creature an opponent controls.
1/2
#It steals fame from an enemy creature.
Chaot Dancer 3R
Creature - Centaur Warrior
Renown 1 (When this creature deals combat damage to a player, if it isn't renowned, put one +1/+1 counter on it and it becomes renowned.)
When Chaot Dancer becomes renowned, it deals 3 damage to target player.
3/3
Bellowing Taitin 3B
Creature - Horror
Renown 1 (When this creature deals combat damage to a player, if it isn't renowned, put one +1/+1 counter on it and it becomes renowned.)
As long as Bellowing Taitin is renowned, it has tromple and flying.
3/3
#Oh! Tromple!! That's too much! My sides!
Dragon's Revelation 2W
Instant
Exile target creature with power 4 or greater. Spell Mastery - If there are two or more instant and/or sorcery cards in your graveyard, then put a +1/+1 counter on each creature you control.
Bool Slash 1UU
Instant
Counter target spell. If that spell is countered this way, exile it with three time counters on it. At the beginning of your upkeep, remove a time counter. When the last is removed, cast it without paying its mana cost. Spell Mastery - If there are two or more instant and/or sorcery cards in your graveyard, instead destroy all creatures. They can't be regenerated.
#Well, that card went from good to incredibly broken very fast.
Wind Purge 3G
Sorcery
Target creature gets +4/+4 until end of turn. Spell Mastery - If there are two or more instant and/or sorcery cards in your graveyard, put a 1/1 white Spirit creature token with flying onto the battlefield.
#Not bad.
Staff of the Medusa W
Enchantment - Aura
Enchant creature Spell Mastery - As long as there are two or more instant and/or sorcery cards in your graveyard, enchanted creature gets +1/+1 for each $TYPE counter on Staff of the Medusa. 3, T: Put a $TYPE counter on Staff of the Medusa. T, Remove any number of $TYPE counters from Staff of the Medusa: You gain 1 life.
#That's... a very elaborate card.
Sword of the Pierced Totem UUU
Enchantment Spell Mastery - As long as there are two or more instant and/or sorcery cards in your graveyard, if you would gain life, draw that many cards instead.
#I'm guessing it was inspired to make a better Lich.
I have some hand-tweaked versions of some of the machine-generated cards I want to post, but I need some help with formatting. How do I make non-standard mana symbols such as Phyrexian, hybrid, etc?
Wind Purge 3G
Sorcery
Target creature gets +4/+4 until end of turn. Spell Mastery - If there are two or more instant and/or sorcery cards in your graveyard, put a 1/1 white Spirit creature token with flying onto the battlefield.
#Not bad.
This is probably getting way ahead of things when the program doesn't yet know that it should combine X costs with X effects and Kicker with Kicked, but... do you think that one day it could learn that generally these Spell Mastery/Threshold/Landfall type effects go from one effect to a larger version of that effect, rather than from one effect to a completely different effect? Like Spell Mastery on Wind Purge upping the bonus to +8/+8, rather than giving you a random token?
That's a problem of a higher order. You'd have to get the reference right, and then you'd have to make it bigger (rather than smaller). Right now I'm working out the coreference issues. If we can get past those more basic issues, then we can consider the "what is the value of this effect" problem. There's a semantic barrier that I hope we can overcome, but I'm still fuzzy on how that's going to work. Possible though? Yes. The network has the wonderful property of being a universal approximator, which means it can learn to do anything if you teach it correctly, lol.
I'm currently generating more data from my existing networks, since I haven't been inspired to train anything new yet. I'm about 40% of the way through generating two 5MB dumps at temperatures 0.5 and 0.8. I have 1MB dumps at each of those temperatures and at 0.3 ready for analysis. I can automatically format and post any amount of that data on the forum, if there's demand for it. Are there any limitations to the amount of text I can post at once? (of course it would be nicely organized in nested spoilers).
A few observations:
First, low temperature seems to interact very badly with my unary encoding. If the temperature is too low, the network will occasionally decide to spit out extremely large numbers. They're so large that it seems to forget what it was doing before it started creating the number - this is the only instance where I've seen it fail to put the closing } on my mana cost format. At temperature 3, it generated at least one number that was many thousands of characters long, so long it gives emacs trouble when I try to load up the file.
Second, the idea of separate X encodings is interesting. This crossed my mind a while ago, though I decided against implementing it. The issue is that the different uses of X are still somewhat similar, so I'm not sure how to do the cost benefit analysis of making things more distinct vs. splitting up an already smallish pool of cards. 300 is not a lot of cards in a pool of 15,000.
I think you might be able to get a bigger benefit removing the ambiguity of the word 'counter,' as in 'counter target spell' vs 'charge counter' or '+1/+1 counter.' I want to hack up my formatting script a little more and implement this tonight, so hopefully I can start training a new network to see if it does anything once my data dumps finish.
PS: Any recommendations for a word to replace 'counter' with? I was thinking of calling it either 'uncast' target spell, or 'cancel' target spell.
I'm currently generating more data from my existing networks, since I haven't been inspired to train anything new yet. I'm about 40% of the way through generating two 5MB dumps at temperatures 0.5 and 0.8. I have 1MB dumps at each of those temperatures and at 0.3 ready for analysis. I can automatically format and post any amount of that data on the forum, if there's demand for it. Are there any limitations to the amount of text I can post at once? (of course it would be nicely organized in nested spoilers).
If it's very large you might do as I did and post a link to a curated text file off of your Google Drive or similar storage.
First, low temperature seems to interact very badly with my unary encoding. If the temperature is too low, the network will occasionally decide to spit out extremely large numbers. They're so large that it seems to forget what it was doing before it started creating the number - this is the only instance where I've seen it fail to put the closing } on my mana cost format. At temperature 3, it generated at least one number that was many thousands of characters long, so long it gives emacs trouble when I try to load up the file.
This humors me greatly. "And add one more mana. And another. And another <10000 iterations later> and done!"
Second, the idea of separate X encodings is interesting. This crossed my mind a while ago, though I decided against implementing it. The issue is that the different uses of X are still somewhat similar, so I'm not sure how to do the cost benefit analysis of making things more distinct vs. splitting up an already smallish pool of cards. 300 is not a lot of cards in a pool of 15,000.
I don't yet have an answer either way, but it'll be on my mind.
I think you might be able to get a bigger benefit removing the ambiguity of the word 'counter,' as in 'counter target spell' vs 'charge counter' or '+1/+1 counter.' I want to hack up my formatting script a little more and implement this tonight, so hopefully I can start training a new network to see if it does anything once my data dumps finish.
PS: Any recommendations for a word to replace 'counter' with? I was thinking of calling it either 'uncast' target spell, or 'cancel' target spell.
Uncast for brevity, neutralize if you're feeling like it should be more latinate.
Alright, if you want to see my files here's a google drive link. That's where I'll be posting my neural net checkpoints of interest and raw output dumps.
I'll be back later with any interesting tidbits and highlights I can find.
Oh, and uncast it is then, unless there are any better suggestions. Maybe the network will even notice that things that can be cast can also be uncast - probably too high of order of an effect to observe, but one can hope.
So I cherry-picked a bunch of cards from the first 10 pages of the thread and, in a lot of cases, tweaked the cards (in some cases, the "tweak" is more like a complete rewrite.) Mana costs, power/toughness, color, keyword abilities, etc. were all subject to adjustment, and in the case of cards that just didn't make any sense as generated, I took liberties to just riff on what the network came up with and let it inspire something original.
Also: I ran a bunch of the nonsense words through Word's spelling suggestions. Sometimes I had to break up longer words and THEN do it before recombining the resulting words. The results are occasionally hilarious such as with the card "Criminally Renamed."
Shring the Artist 2BB
Legendary Creature - Cat
Flying
Whenever you cast a spell, you may return target creature card from your graveyard to your hand.
2/2
Legan of Echince 7GG
Legendary Creature - Treefolk Avatar
Treefolk you control have Forestwalk.
At the beginning of your upkeep, put a +1/+1 counter on target creature. GGG: Target land becomes a Forest in addition to its other types (this effect lasts indefinitely).
6/5
Profile Shield BR
Creature - Skeleton
Haste
When $THIS leaves the battlefield, sacrifice it and each opponent discards a card.
2/2
Minded Hired Pariah’s Scepter 3
Artifact 2(W/P), T: Put a 1/1 white Advisor creature token onto the battlefield. It has “This creature can’t be blocked by creatures with power greater than 2.”
Slothward Bastrop 3
Artifact 2(B/P),t: Return target creature card from your graveyard to your hand.
Anabolic Cobble 3
Artifact 2(U/P),T: Target land becomes a 4/4 blue Elemental creature with flying until end of turn.
Volleying Stall 3
Artifact 2(G/P):,T: Put a 1/1 green Saproling creature token onto the battlefield. It has “Sacrifice this creature: prevent all combat damage target creature would deal this turn.”
Blood Rasht 1R
Creature - Goblin Shaman 3R, sacrifice a land: $THIS deals 3 damage to target player.
2/2
Creesavage Ally 2WW
Creature - Elemental Wizard
Flying
Cycling W
When you cycle $THIS, prevent the next 2 damage that would be dealt to you this turn.
3/3
Prochonomancer 6
Artifact Creature – Golem Wizard
Whenever $THIS attacks or blocks alone, you may pay 1R. If you do, put a +1/+1 counter on it.
5/5
Carnot Sprat 3BB
Creature - Insect
Deathtouch
When $THIS deals combat damage to a player, you may put a 2/2 black Insect creature token with deathtouch onto the battlefield.
2/2
Strong Shield 3B
Creature - Zombie Knight
Whenever $THIS attacks, you may return target creature card from your graveyard to your hand.
2/2
Uniclings 2GG
Creature - Beast GGG,T, Sacrifice $THIS: Choose one -
* Destroy all artifacts.
* Put a +1/+1 counter on each creature you control.
2/2
Schengen Helldandies 1BG
Creature – Zombie Elf 0: Add G to your mana pool. If you do, put a -1/-1 counter on $THIS. Play this ability no more than once each turn.
3/4
Arnostune UUB
Instant
Kicker 1
Buyback 1
Scry 1
If $THIS was kicked, scry 2 instead.
Handset 2U
Counter target spell. If you do, its controller may put a +1/+1 counter on a creature he or she controls.
Invader Cycloid 2B
Creature - Spirit B, sacrifice a creature: Return target creature card with converted mana cost 4 or less from your graveyard to your hand. You gain life equal to the sacrificed creature's converted mana cost.
2/2
Retile Volant W
Creature - Human Knight
Whenever $THIS blocks or becomes blocked, you may pay 1W. If you do, prevent all combat damage that would be dealt to and by $THIS this turn and you gain 2 life.
1/1
Bornfesters Kinasoressurge
Land T: Add 1 to your mana pool. T: Destroy target creature without any keyword abilities.
Revenging Slayer 3UUB
Legendary Creature — Human Wizard
Sacrifice a creature: Regenerate target artifact. Target player loses 1 life and you gain 1 life.
3/4
Sacsexon Dose 4UU
Creature - Dragon
Evoke 2UU
Flying
When $THIS enters the battlefield, creatures you control gain hexproof and protection from red and from green until end of turn.
4/4
Wahid Gorged Havoc 2R
Creature - Drgon
Flying
Discard a card, tap an untapped basic land you control: $THIS deals 2 damage to target creature or player.
1/7
Retaken Searcher 3UU
Creature – Human Wizard
Flying, Hexproof
Whenever $THIS deals combat damage to an opponent, you may have it become a copy of target creature. If you do, search your library for a card named $THIS and put it into your hand. Then shuffle your library.
2/2
Theron Deric 2WW
Legendary Creature - Human Cleric
Vigilance
Morph 4WW (You may cast this card face down as a 2/2 creature for 3. Turn it face up any time for its morph cost.)
When $THIS is turned face up, it gains indestructible until end of turn and you redirect all damage that would be dealt to you this turn to $THIS. If the total damage redirected to $THIS this way is equal to or greater than your life total at the time it was turned face up, sacrifice it.
2/4
Sparktuner’s Hirsh 2R
Creature - Elemental
Flying, trample
When $THIS enters the battlefield, it gains haste until end of turn. R: $THIS gets +1/+0 until end of turn.
0/3
Charred Marauder 1BB
Creature – Zombie Knight
Intimidate
At the beginning of your upkeep, draw a card, then discard a card.
2/2
Rampager Kinds 1U
Creature – Human Advisor
Flying
Morph 2UU (You may cast this card face down as a 2/2 creature for 3. Turn it face up any time for its morph cost.)
When $THIS is turned face up, put into a graveyard from the battlefield, or a legendary creature enters the battlefield under your control, look at the top seven cards of your library. Exile any number of those cards, then put the rest back in any order.
1/1
Sorcerer Hellking 2(U/P)R
Legendary Creature – Dragon 2R: $THIS deals 1 damage to target creature or player. Prevent the next 1 damage that would be dealt to each artifact creature you control.
Fading 12
Remove a fade counter from $THIS: Add 3 to your mana pool. Spend this mana only to cast artifact creature spells and no more than 3 per spell. The next artifact creature you cast comes into play with fading 1 and gains haste until end of turn.
4/3
Crum Choler 1WW
Enchantment
Whenever you cast a creature spell, tap target creature with power 3 or greater.
Giantblood Screecher
Legendary Land T: Add R to your mana pool. 1RR, T: Target land you control becomes a 3/3 Elemental creature that's still a land. Any player may pay 2. If a player does, $THIS deals 3 damage to that player and you exile that land at the beginning of the next end step.
Simic Key 0
Artifact
$THIS enters the battlefield tapped. T: Add U or G to your mana pool. Spend this mana only to cast creature spells.
Anahi of Vine Disciple
Legendary Land
$THIS enters the battlefield tapped. T: Add RRR to your mana pool. $THIS deals 3 damage to target creature you own and 3 damage to that creature's controller.
Divinity 3B
Creature – Zombie Crocodile 2B, Sacrifice a creature: $THIS deals 3 damage to target creature or player. If the target of this ability is a green creature, $THIS gets +1/+2 until end of turn.
3/3
Martel Savage RR
Instant
$THIS deals 1 damage to each creature.
Salvage 2RR (2RR, Exile this card from your graveyard: put a number of +1/+1 counters on target creature equal to this card’s converted mana cost. Salvage only as a sorcery.)
Illusionate BU
Instant
Destroy target creature. It can’t be regenerated. At end of turn, its owner may return it from the graveyard to the battlefield.
Frostwreath 1U
Enchantment – Aura
Enchant creature
Fading 2
You control enchanted creature. U: Enchanted creature gets +1/+0 until end of turn.
Flocky of the Dust 2RG
Creature – Werewolf
Haste
Fading 3
$THIS gets +1/+1 for each fade counter on it
Whenever you remove a fade counter from $THIS, put a +1/+1 counter on another target creature. That creature becomes a Werewolf in addition to its other types.
4/3
Mortenish Sather 3RG
Legendary Creature – Elf Rogue
Haste
Whenever another creature enters the battlefield under your control, $THIS deals 2 damage to target creature or player.
3/3
Barren Mudroom 1BB
Sorcery
You may exile any number of cards from your graveyard. If you do, return up to X creature cards from your graveyard to your hand, where X is the number of cards exiled this way.
Marek Head 2G
Creature - Elf Rogue
Reach GG: $THIS deals 1 damage to target attacking or blocking creature with flying. That creature loses flying until end of turn.
1/3
Candled 2G
Instant
Target creature gets +2/+5 and gains hexproof until end of turn.
Cogan Attar 1RR
Legendary Creature - Goblin Rogue
Protection from blue
$THIS can’t be countered.
When $THIS dies, destroy target creature and $THIS deals 3 damage to each player.
3/3
Criminally Renamed 2UU
Creature – Merfolk Wizard
Whenever $THIS becomes untapped, you may draw a card.
3/3
Whited Devotees 4WW
Legendary Creature - Human Knight
When $THIS enters the battlefield, exile all creatures target player controls with power 2 or greater.
4/4
Cartel Nightmares RW
Instant
Target creature gets +2/+2 and gains haste and vigilance until end of turn.
Flashback 4RW
Hunt-Torero UBR
Legendary Creature - Merfolk Wizard
Hexproof
When $THIS dies, it deals 3 damage to target creature or player.
Whenever a creature dealt combat damage by $THIS this turn dies, $THIS deals damage equal to that creature's power to target creature or player. BR,, Sacrifice a creature: $THIS deals damage to target creature or player equal to the sacrificed creature's toughness. BB: Return $THIS from your graveyard to your hand.
2/2
Morrow Hackly-Cut 3WG
Legendary Creature – Elemental Spirit
Flying, vigilance T: Add G to your mana pool. T: Add one mana of any color to your mana pool for each other attacking or blocking creature. W, T: Exile target attacking or blocking creature.
3/3
Shriekcaller's Coatls 2GW
Enchantment
Attacking creatures get +1/+1, flying, first strike, and trample.
Conatus Cagers 2WW
Creature – Human Knight
Whenever $THIS attacks, destroy all creatures with power 4 or greater.
2/2
Spires of Impaltros 2R
Legendary Creature - Elder Dreatcan
When $THIS dies, return it to the battlefield tapped under its owner's control.
2/2
Ancient Dodger U
Creature - Human Wizard T: Regenerate target artifact.
1/1
Scion of Ecguiten G
Creature - Cat Warrior
Whenever another creature dies, you may draw a card.
1/1
Total Field Death 3B
Creature – Vampire
Wither
$THIS enters the battlefield with four +1/+1 counters on it.
Whenever $THIS damages a creature, that creature becomes a Vampire in addition to its other types as long as it has any -1/-1 counters on it. 1BB, remove a +1/+1 counter from $THIS: Put a +1/+1 counter on target creature. It becomes a Vampire in addition to its other types. 1BB, {T}: Gain control of target Vampire with at least one +1/+1 or -1/-1 counter on it for as long as you control $THIS.
0/0
Thru 1U
Instant
Target player reveals his or her hand. Choose a card in that player’s hand. Counter target spell if it has the same name as that card.
Draw a card.
Cribstalker Giant 3B
Creature – Zombie Giant Assassin
When $THIS dies, return it to its owner's hand.
2/2
Blood Boar 1R
Creature - Beast
Haste
Players can't gain life. Damage can’t be prevented.
2/1
Ahhhhh Bellowing Taitin, Tromple is back, yes, I can't handle this!
Also is the first part of Bool Slash copied from an existing card, or is it a new design? Cos it's definitely fun. (Counter target spell. If that spell is countered this way, exile it with three time counters on it. At the beginning of your upkeep, remove a time counter. When the last is removed, cast it without paying its mana cost.)
My suggestion for 'uncast' is 'nope'. As in "Nope target spell."
Mustard: Bool Slash is Delay from Time Spiral block, with upside.
Just wanted to say that this thread is awesome, and I would like to point out that the above card Dragon's Revelation is a completely reasonable card that could actually be in Origins.
Sword of the Pierced Totem is actually completely feasible as well. Well, except for the minor point that I don't think Spell Mastery shows up on non-Sorcery/Instant spells.
I am really looking forward to a created set of cards with the network.
Talcos, I am correct in believing that this will be able to (at some point) take a design skeleton and mechanics for a set and create the set, correct?
Yeah I figured it had to be real but my search failed to find it, thanks.
Sorcerer Hellking above has some weird mana cost issues in a couple of places. And some other cards still seem cheap. But it was fun to read, Xenus_Paradox.
Yeah I figured it had to be real but my search failed to find it, thanks.
Sorcerer Hellking above has some weird mana cost issues in a couple of places. And some other cards still seem cheap. But it was fun to read, Xenus_Paradox.
Damn curly brackets. The're real particular about where you put 'em. Should be working now.
I haven't played the game in years, though I do check in to see what's going on with it once in a while, so fine balance isn't necessarily my strongest suit.
Private Mod Note
():
Rollback Post to RevisionRollBack
To post a comment, please login or register a new account.
"if you have no cards in hand, you may put a creature card from your hand onto the battlefield"
You may, but you can't
So far, my girlfriend's favorite card name that my RNN has generated is "Morti****er."
At temperature 0.5, that's very good, honestly, and that brings up an interesting point. At lower temperatures, I have the feeling that the network is more reliably producing cards with abilities and effects that are on point in terms of color. Of course, that's just from what I see, I'm sure we can measure the in some way to confirm that. Still, that's very promising. I'll have to write a script later for mid-sequence priming based on the input format. I'm very curious to see what multicolored creatures look like. Since they're rarer occurrences, it might just give a hazy mix of multicolored text that may or may not be relevant to the creatures colors. For example, if I do primetext "|crafty merfolk||creature||merfolk rogue|&^^/&^^|{^UG}|" and I sample at temperature 0.5, I can get cards like
Crafty Merfolk
UG
Crafty Merfolk can't be blocked.
When Crafty Merfolk enters the battlefield, sacrifice it unless you sacrifice a black or red permanent.
2/2
Emphasis mine. Notice that you can get cards that do multicolored things but the colors involved aren't necessarily the colors of the creature.
On the other hand, it's difficult to track what the network is thinking because if you prime with "|mistfire blade||creature||human warrior spellshaper|&^^^/&^|{^^^RU}" you can also get cards like
Mistfire Blade
3UR
Creature - Human Warrior
2UU, T, discard a card: draw a card.
3/1
Which is fine for a red creature. However, if I prime it with the colors in the mana cost reversed (blue after red), with the same seed, temperature, and everything, I get...
Mistfire Blade
3RU
Creature - Human Warrior
3, T: Target creature gains flying until end of turn.
3/1
We get an effect that is more blue than red, which implies that the network thinks that the last mana symbol, which is closest to the body text, is most important to determining what should be written in the card. Since the majority of creatures are monocolored, that rule makes perfect sense. The network usually converges on the least complicated rule for predicting the text, so if it can get away with counting the number of mana symbols (how big or splashy the card needs to be) and the type of the last mana symbol (what color the card needs to be), then it's going to do that.
The trick then, I feel, is more robust input generation for the training process. We have to force it into situations where it gets the card wrong unless it takes all of the colors into consideration. I think we can do that.
For those of you who want a comparison in recent literature, there's Zaremba and Sutskever 2015. In that paper, the authors teach a 2 layer LSTM network with 400 neurons per layer that reads in characters one at a time (in short, much like ours) to simulate the execution of Python code. The authors found that the secret to getting it to understand Python code is to feed the network simple examples of generated programs and then build up to more complicated generated ones, a strategy called curriculum learning.
Following with that reasoning, I previously implemented a generator earlier for simple creatures to help the network grasp relationships between mana costs, abilities, and power and toughness, and I was reasonably successful with that. I think the next step is to extend that type of approach to more card types, multicolored cards, etc.
With a good encoding (for which we have several good candidates), good card generation (for which we have some work done already), and a good training curriculum (for which we have support in the literature), we should be able to up the quality of the average card produced by the network considerably. From there, careful application of script-driven priming should get us the rest of the way to having full-fledged sets. That's the trajectory I have in mind.
I'm going to have so much fun drafting these cards.
My LinkedIn profile... thing (I have one of those now!).
My research team's webpage.
The mtg-rnn repo and the mtg-encode repo.
This guy is actually super neat, although would work better as a bigger creature that taps all other creatures. I like that within all the random junk there are tiny gems of elegance like this one.
"A Plague on All Your Houses!" - Thespian's Stage Pox
The problem is related to coreference resolution. For example, say you have the following English sentence:
"John left the house. He went straight to work."
"John" and "He" are coreferents, in this case, "John" is the antecedent which defines "He". You can have the first sentence without the second and have everything be syntactically and semantically fine, but the second sentence is missing semantic information without the first (who does "He" refer to?).
There are many examples of coreference in Magic text, here are some examples:
* "Counter target spell. Its controller may draw up to two cards at the beginning of the next turn's upkeep."
* "Kicker 3 ... If Æther Figment was kicked, it enters the battlefield with two +1/+1 counters on it."
* "Enchant creature ... Enchanted creature gets +1/+2"
Now, the network has to learn from scratch, with no prior knowledge about what the text means or what relationships exist. Now, it does a decent job of picking up on those kinds of relationships when it has lots of examples, like aura cards or equipment cards. It sometimes falls short on others:
And this same problem is showing up with X costs. These are especially tough situations where you have two coreferents that depend upon each other to be meaningful. Of course, it's fixable with the right training.
My LinkedIn profile... thing (I have one of those now!).
My research team's webpage.
The mtg-rnn repo and the mtg-encode repo.
With kicker, that makes some sense, as a simple gatherer search for "kicker" returned only 113 cards that have kicker (this includes multi-kicker). There are 236 cards with X in the mana cost, and many more that have both occurrences in the rules text. I would have expected there to be enough X cards for it to figure it out. In any case, generating large amounts of cards with kicker (and other abilities that have this coreferent problem) seems to be the way to deal with this. I suppose the trick to that is making sure those 'training' cards aren't skewing the power level for cards with that ability.
My thought exactly. For example, it's fine with generating cards that have kicker and it's fine with generating cards that have "if kicked" clauses, we just need to teach it you can't have one without the other, that there is some kind of mutual dependency between them. Feeding it additional generated cards could solve that problem.
Heck, we can generate those cards with priming, actually. If I prime the card to generate "Kicker <cost>" followed by an "if kicked clause", it can fill in what happens when the conditions are met. That may be one way of training it to recognize that relationship, with a feedback loop at low temperature. I might try that in the near future and see what I get.
My LinkedIn profile... thing (I have one of those now!).
My research team's webpage.
The mtg-rnn repo and the mtg-encode repo.
Indeed, and unless I'm mistaken, I am already doing this, because the text is all lowercase. The only uppercase letters are for mana symbols and occurrences of X, which looks the same inside and outside of mana symbols (though to be consistent with other mana symbols, X in a mana symbol is actually written XX).
(also hello I've been lurking the thread for some time but for whatever reason this is the first time I'm moved to comment)
ETA: also from h_s's cards wow, that network sure has a thing for extremely expensive black sorceries doesn't it?
X = X as a mana cost, and abilities, counters that are directly related from X mana being paid in the casting cost.
Example: Fireball deals X damage, or Creature comes into play with X +1/+1 counters on it.
XA : X as a paid ability, when the casting cost of the card has a set converted mana cost.
Example: Creature costs 1MM, has ability "Pay XA: Sacrifice an artifact, ThisCard deals XA damage to target creature or player."
XY = X is dependent on a secondary value or condition.
Example: Creature enters the battlefield with XY +1/+1 counters on it where XY is the number of cards in your hand.
HAHAHAHA
Can this thread just be stickied already
That's a possibility, considering X can be used in multiple contexts. It would definitely lessen the cognitive burden on the network, which makes it more likely to learn the relationship. Still, if we make more progress on the "kicker/kicked" front, that should ameliorate the quality of our results. Or so I hope, I'm really just hypothesizing at this point until I get more data to back it up... and data I shall have, soon.
I'm glad you enjoy my sense of humor, lol.
EDIT: I was having some fun just a moment ago...
Loyal Hornemaker
1RR
Creature - Goblin Artificer
Renown 2 (When this creature deals combat damage to a player, if it isn't renowned, put two +1/+1 counters on it and it becomes renowned.)
When Loyal Hornemaker becomes renowned, remove a +1/+1 counter from it.
1/1
#Not quite what I had in mind, but at least it understood that Loyal Hornemaker could have counters to take away.
Hardowed Healer
3UU
Creature - Spirit
Renown 2 (When this creature deals combat damage to a player, if it isn't renowned, put two +1/+1 counters on it and it becomes renowned.)
When Hardowed Healer becomes renowned, remove all +1/+1 counters from target creature an opponent controls.
1/2
#It steals fame from an enemy creature.
Chaot Dancer
3R
Creature - Centaur Warrior
Renown 1 (When this creature deals combat damage to a player, if it isn't renowned, put one +1/+1 counter on it and it becomes renowned.)
When Chaot Dancer becomes renowned, it deals 3 damage to target player.
3/3
Bellowing Taitin
3B
Creature - Horror
Renown 1 (When this creature deals combat damage to a player, if it isn't renowned, put one +1/+1 counter on it and it becomes renowned.)
As long as Bellowing Taitin is renowned, it has tromple and flying.
3/3
#Oh! Tromple!! That's too much! My sides!
Dragon's Revelation
2W
Instant
Exile target creature with power 4 or greater.
Spell Mastery - If there are two or more instant and/or sorcery cards in your graveyard, then put a +1/+1 counter on each creature you control.
Bool Slash
1UU
Instant
Counter target spell. If that spell is countered this way, exile it with three time counters on it. At the beginning of your upkeep, remove a time counter. When the last is removed, cast it without paying its mana cost.
Spell Mastery - If there are two or more instant and/or sorcery cards in your graveyard, instead destroy all creatures. They can't be regenerated.
#Well, that card went from good to incredibly broken very fast.
Wind Purge
3G
Sorcery
Target creature gets +4/+4 until end of turn.
Spell Mastery - If there are two or more instant and/or sorcery cards in your graveyard, put a 1/1 white Spirit creature token with flying onto the battlefield.
#Not bad.
Staff of the Medusa
W
Enchantment - Aura
Enchant creature
Spell Mastery - As long as there are two or more instant and/or sorcery cards in your graveyard, enchanted creature gets +1/+1 for each $TYPE counter on Staff of the Medusa.
3, T: Put a $TYPE counter on Staff of the Medusa.
T, Remove any number of $TYPE counters from Staff of the Medusa: You gain 1 life.
#That's... a very elaborate card.
Sword of the Pierced Totem
UUU
Enchantment
Spell Mastery - As long as there are two or more instant and/or sorcery cards in your graveyard, if you would gain life, draw that many cards instead.
#I'm guessing it was inspired to make a better Lich.
My LinkedIn profile... thing (I have one of those now!).
My research team's webpage.
The mtg-rnn repo and the mtg-encode repo.
I have some hand-tweaked versions of some of the machine-generated cards I want to post, but I need some help with formatting. How do I make non-standard mana symbols such as Phyrexian, hybrid, etc?
I think, [mana]{Pb}[/mana] is (P/B), and [mana]{g/w}[/mana] is (G/W).
Or not.. what IS the code for phyrexian mana symbols? I'll have to check.
My LinkedIn profile... thing (I have one of those now!).
My research team's webpage.
The mtg-rnn repo and the mtg-encode repo.
Just flip the P and b... so {bp}
Choose one of these judge of creation:
Make Strionic Resonator shine!
You can not grasp the true form of Ashiok's attack!
That's a problem of a higher order. You'd have to get the reference right, and then you'd have to make it bigger (rather than smaller). Right now I'm working out the coreference issues. If we can get past those more basic issues, then we can consider the "what is the value of this effect" problem. There's a semantic barrier that I hope we can overcome, but I'm still fuzzy on how that's going to work. Possible though? Yes. The network has the wonderful property of being a universal approximator, which means it can learn to do anything if you teach it correctly, lol.
My LinkedIn profile... thing (I have one of those now!).
My research team's webpage.
The mtg-rnn repo and the mtg-encode repo.
A few observations:
First, low temperature seems to interact very badly with my unary encoding. If the temperature is too low, the network will occasionally decide to spit out extremely large numbers. They're so large that it seems to forget what it was doing before it started creating the number - this is the only instance where I've seen it fail to put the closing } on my mana cost format. At temperature 3, it generated at least one number that was many thousands of characters long, so long it gives emacs trouble when I try to load up the file.
Second, the idea of separate X encodings is interesting. This crossed my mind a while ago, though I decided against implementing it. The issue is that the different uses of X are still somewhat similar, so I'm not sure how to do the cost benefit analysis of making things more distinct vs. splitting up an already smallish pool of cards. 300 is not a lot of cards in a pool of 15,000.
I think you might be able to get a bigger benefit removing the ambiguity of the word 'counter,' as in 'counter target spell' vs 'charge counter' or '+1/+1 counter.' I want to hack up my formatting script a little more and implement this tonight, so hopefully I can start training a new network to see if it does anything once my data dumps finish.
PS: Any recommendations for a word to replace 'counter' with? I was thinking of calling it either 'uncast' target spell, or 'cancel' target spell.
If it's very large you might do as I did and post a link to a curated text file off of your Google Drive or similar storage.
This humors me greatly. "And add one more mana. And another. And another <10000 iterations later> and done!"
I don't yet have an answer either way, but it'll be on my mind.
Uncast for brevity, neutralize if you're feeling like it should be more latinate.
My LinkedIn profile... thing (I have one of those now!).
My research team's webpage.
The mtg-rnn repo and the mtg-encode repo.
I'll be back later with any interesting tidbits and highlights I can find.
Oh, and uncast it is then, unless there are any better suggestions. Maybe the network will even notice that things that can be cast can also be uncast - probably too high of order of an effect to observe, but one can hope.
Also: I ran a bunch of the nonsense words through Word's spelling suggestions. Sometimes I had to break up longer words and THEN do it before recombining the resulting words. The results are occasionally hilarious such as with the card "Criminally Renamed."
Shring the Artist
2BB
Legendary Creature - Cat
Flying
Whenever you cast a spell, you may return target creature card from your graveyard to your hand.
2/2
Legan of Echince
7GG
Legendary Creature - Treefolk Avatar
Treefolk you control have Forestwalk.
At the beginning of your upkeep, put a +1/+1 counter on target creature.
GGG: Target land becomes a Forest in addition to its other types (this effect lasts indefinitely).
6/5
Profile Shield
BR
Creature - Skeleton
Haste
When $THIS leaves the battlefield, sacrifice it and each opponent discards a card.
2/2
Minded Hired Pariah’s Scepter
3
Artifact
2(W/P), T: Put a 1/1 white Advisor creature token onto the battlefield. It has “This creature can’t be blocked by creatures with power greater than 2.”
Slothward Bastrop
3
Artifact
2(B/P),t: Return target creature card from your graveyard to your hand.
Anabolic Cobble
3
Artifact
2(U/P),T: Target land becomes a 4/4 blue Elemental creature with flying until end of turn.
Volleying Stall
3
Artifact
2(G/P):,T: Put a 1/1 green Saproling creature token onto the battlefield. It has “Sacrifice this creature: prevent all combat damage target creature would deal this turn.”
Blood Rasht
1R
Creature - Goblin Shaman
3R, sacrifice a land: $THIS deals 3 damage to target player.
2/2
Creesavage Ally
2WW
Creature - Elemental Wizard
Flying
Cycling W
When you cycle $THIS, prevent the next 2 damage that would be dealt to you this turn.
3/3
Prochonomancer
6
Artifact Creature – Golem Wizard
Whenever $THIS attacks or blocks alone, you may pay 1R. If you do, put a +1/+1 counter on it.
5/5
Carnot Sprat
3BB
Creature - Insect
Deathtouch
When $THIS deals combat damage to a player, you may put a 2/2 black Insect creature token with deathtouch onto the battlefield.
2/2
Strong Shield
3B
Creature - Zombie Knight
Whenever $THIS attacks, you may return target creature card from your graveyard to your hand.
2/2
Uniclings
2GG
Creature - Beast
GGG,T, Sacrifice $THIS: Choose one -
* Destroy all artifacts.
* Put a +1/+1 counter on each creature you control.
2/2
Schengen Helldandies
1BG
Creature – Zombie Elf
0: Add G to your mana pool. If you do, put a -1/-1 counter on $THIS. Play this ability no more than once each turn.
3/4
Arnostune
UUB
Instant
Kicker 1
Buyback 1
Scry 1
If $THIS was kicked, scry 2 instead.
Handset
2U
Counter target spell. If you do, its controller may put a +1/+1 counter on a creature he or she controls.
Invader Cycloid
2B
Creature - Spirit
B, sacrifice a creature: Return target creature card with converted mana cost 4 or less from your graveyard to your hand. You gain life equal to the sacrificed creature's converted mana cost.
2/2
Retile Volant
W
Creature - Human Knight
Whenever $THIS blocks or becomes blocked, you may pay 1W. If you do, prevent all combat damage that would be dealt to and by $THIS this turn and you gain 2 life.
1/1
Bornfesters Kinasoressurge
Land
T: Add 1 to your mana pool.
T: Destroy target creature without any keyword abilities.
Revenging Slayer 3UUB
Legendary Creature — Human Wizard
Sacrifice a creature: Regenerate target artifact. Target player loses 1 life and you gain 1 life.
3/4
Fountain Despair
2BB
Creature - Vampire
Flying, haste
3/2
Sacsexon Dose
4UU
Creature - Dragon
Evoke 2UU
Flying
When $THIS enters the battlefield, creatures you control gain hexproof and protection from red and from green until end of turn.
4/4
Wahid Gorged Havoc
2R
Creature - Drgon
Flying
Discard a card, tap an untapped basic land you control: $THIS deals 2 damage to target creature or player.
1/7
Retaken Searcher
3UU
Creature – Human Wizard
Flying, Hexproof
Whenever $THIS deals combat damage to an opponent, you may have it become a copy of target creature. If you do, search your library for a card named $THIS and put it into your hand. Then shuffle your library.
2/2
Theron Deric
2WW
Legendary Creature - Human Cleric
Vigilance
Morph 4WW (You may cast this card face down as a 2/2 creature for 3. Turn it face up any time for its morph cost.)
When $THIS is turned face up, it gains indestructible until end of turn and you redirect all damage that would be dealt to you this turn to $THIS. If the total damage redirected to $THIS this way is equal to or greater than your life total at the time it was turned face up, sacrifice it.
2/4
Sparktuner’s Hirsh
2R
Creature - Elemental
Flying, trample
When $THIS enters the battlefield, it gains haste until end of turn.
R: $THIS gets +1/+0 until end of turn.
0/3
Charred Marauder 1BB
Creature – Zombie Knight
Intimidate
At the beginning of your upkeep, draw a card, then discard a card.
2/2
Rampager Kinds 1U
Creature – Human Advisor
Flying
Morph 2UU (You may cast this card face down as a 2/2 creature for 3. Turn it face up any time for its morph cost.)
When $THIS is turned face up, put into a graveyard from the battlefield, or a legendary creature enters the battlefield under your control, look at the top seven cards of your library. Exile any number of those cards, then put the rest back in any order.
1/1
Sorcerer Hellking 2(U/P)R
Legendary Creature – Dragon
2R: $THIS deals 1 damage to target creature or player. Prevent the next 1 damage that would be dealt to each artifact creature you control.
Fading 12
Remove a fade counter from $THIS: Add 3 to your mana pool. Spend this mana only to cast artifact creature spells and no more than 3 per spell. The next artifact creature you cast comes into play with fading 1 and gains haste until end of turn.
4/3
Crum Choler 1WW
Enchantment
Whenever you cast a creature spell, tap target creature with power 3 or greater.
Giantblood Screecher
Legendary Land
T: Add R to your mana pool.
1RR, T: Target land you control becomes a 3/3 Elemental creature that's still a land. Any player may pay 2. If a player does, $THIS deals 3 damage to that player and you exile that land at the beginning of the next end step.
Simic Key 0
Artifact
$THIS enters the battlefield tapped.
T: Add U or G to your mana pool. Spend this mana only to cast creature spells.
Anahi of Vine Disciple
Legendary Land
$THIS enters the battlefield tapped.
T: Add RRR to your mana pool. $THIS deals 3 damage to target creature you own and 3 damage to that creature's controller.
Divinity 3B
Creature – Zombie Crocodile
2B, Sacrifice a creature: $THIS deals 3 damage to target creature or player. If the target of this ability is a green creature, $THIS gets +1/+2 until end of turn.
3/3
Martel Savage RR
Instant
$THIS deals 1 damage to each creature.
Salvage 2RR (2RR, Exile this card from your graveyard: put a number of +1/+1 counters on target creature equal to this card’s converted mana cost. Salvage only as a sorcery.)
Illusionate BU
Instant
Destroy target creature. It can’t be regenerated. At end of turn, its owner may return it from the graveyard to the battlefield.
Frostwreath
1U
Enchantment – Aura
Enchant creature
Fading 2
You control enchanted creature.
U: Enchanted creature gets +1/+0 until end of turn.
Flocky of the Dust
2RG
Creature – Werewolf
Haste
Fading 3
$THIS gets +1/+1 for each fade counter on it
Whenever you remove a fade counter from $THIS, put a +1/+1 counter on another target creature. That creature becomes a Werewolf in addition to its other types.
4/3
Mortenish Sather
3RG
Legendary Creature – Elf Rogue
Haste
Whenever another creature enters the battlefield under your control, $THIS deals 2 damage to target creature or player.
3/3
Barren Mudroom
1BB
Sorcery
You may exile any number of cards from your graveyard. If you do, return up to X creature cards from your graveyard to your hand, where X is the number of cards exiled this way.
Marek Head
2G
Creature - Elf Rogue
Reach
GG: $THIS deals 1 damage to target attacking or blocking creature with flying. That creature loses flying until end of turn.
1/3
Candled
2G
Instant
Target creature gets +2/+5 and gains hexproof until end of turn.
Nobelium Rids
6
Legendary Artifact Creature - Golem
Creatures can't block.
13/3
Cogan Attar
1RR
Legendary Creature - Goblin Rogue
Protection from blue
$THIS can’t be countered.
When $THIS dies, destroy target creature and $THIS deals 3 damage to each player.
3/3
Criminally Renamed
2UU
Creature – Merfolk Wizard
Whenever $THIS becomes untapped, you may draw a card.
3/3
Whited Devotees
4WW
Legendary Creature - Human Knight
When $THIS enters the battlefield, exile all creatures target player controls with power 2 or greater.
4/4
Cartel Nightmares
RW
Instant
Target creature gets +2/+2 and gains haste and vigilance until end of turn.
Flashback 4RW
Hunt-Torero
UBR
Legendary Creature - Merfolk Wizard
Hexproof
When $THIS dies, it deals 3 damage to target creature or player.
Whenever a creature dealt combat damage by $THIS this turn dies, $THIS deals damage equal to that creature's power to target creature or player.
BR,, Sacrifice a creature: $THIS deals damage to target creature or player equal to the sacrificed creature's toughness.
BB: Return $THIS from your graveyard to your hand.
2/2
Morrow Hackly-Cut
3WG
Legendary Creature – Elemental Spirit
Flying, vigilance
T: Add G to your mana pool.
T: Add one mana of any color to your mana pool for each other attacking or blocking creature.
W, T: Exile target attacking or blocking creature.
3/3
Shriekcaller's Coatls
2GW
Enchantment
Attacking creatures get +1/+1, flying, first strike, and trample.
Conatus Cagers
2WW
Creature – Human Knight
Whenever $THIS attacks, destroy all creatures with power 4 or greater.
2/2
Spires of Impaltros
2R
Legendary Creature - Elder Dreatcan
When $THIS dies, return it to the battlefield tapped under its owner's control.
2/2
Ancient Dodger
U
Creature - Human Wizard
T: Regenerate target artifact.
1/1
Scion of Ecguiten
G
Creature - Cat Warrior
Whenever another creature dies, you may draw a card.
1/1
Total Field Death
3B
Creature – Vampire
Wither
$THIS enters the battlefield with four +1/+1 counters on it.
Whenever $THIS damages a creature, that creature becomes a Vampire in addition to its other types as long as it has any -1/-1 counters on it.
1BB, remove a +1/+1 counter from $THIS: Put a +1/+1 counter on target creature. It becomes a Vampire in addition to its other types.
1BB, {T}: Gain control of target Vampire with at least one +1/+1 or -1/-1 counter on it for as long as you control $THIS.
0/0
Thru
1U
Instant
Target player reveals his or her hand. Choose a card in that player’s hand. Counter target spell if it has the same name as that card.
Draw a card.
Cribstalker Giant
3B
Creature – Zombie Giant Assassin
When $THIS dies, return it to its owner's hand.
2/2
Blood Boar
1R
Creature - Beast
Haste
Players can't gain life. Damage can’t be prevented.
2/1
Also is the first part of Bool Slash copied from an existing card, or is it a new design? Cos it's definitely fun. (Counter target spell. If that spell is countered this way, exile it with three time counters on it. At the beginning of your upkeep, remove a time counter. When the last is removed, cast it without paying its mana cost.)
My suggestion for 'uncast' is 'nope'. As in "Nope target spell."
Just wanted to say that this thread is awesome, and I would like to point out that the above card Dragon's Revelation is a completely reasonable card that could actually be in Origins.
Sword of the Pierced Totem is actually completely feasible as well. Well, except for the minor point that I don't think Spell Mastery shows up on non-Sorcery/Instant spells.
I am really looking forward to a created set of cards with the network.
Talcos, I am correct in believing that this will be able to (at some point) take a design skeleton and mechanics for a set and create the set, correct?
Sorcerer Hellking above has some weird mana cost issues in a couple of places. And some other cards still seem cheap. But it was fun to read, Xenus_Paradox.
Damn curly brackets. The're real particular about where you put 'em. Should be working now.
I haven't played the game in years, though I do check in to see what's going on with it once in a while, so fine balance isn't necessarily my strongest suit.