Oh, this is too good not to mention. I coordinated with Mtgsalvation member pickfifteen who has created a twitter account that posts automatically generated renders of cards created with the RNN network. I look forward to seeing what it produces.
EDIT: Also, Izzy, very excited to see what you come up with. My vote's on Shring the Artist or Slidshocking Krow. But maybe it's not fair for me to vote, since they're the ones created by my own creation, haha.
Hey everyone! I was linked to this awesome topic on Facebook by my friend, Andy.
My name is Izzy and I do art for Magic!
I thought it might be fun if you guys pick one of these generated card concepts and I'd do a quick illustration for it!
If you all would like to vote for your favorite and someone do me a favor and tally them up, I'll take the chosen concept
and do it's card art, just like we do for real Magic cards!
Hey everyone! I was linked to this awesome topic on Facebook by my friend, Andy.
My name is Izzy and I do art for Magic!
I thought it might be fun if you guys pick one of these generated card concepts and I'd do a quick illustration for it!
If you all would like to vote for your favorite and someone do me a favor and tally them up, I'll take the chosen concept
and do it's card art, just like we do for real Magic cards!
Whatever you come up with, I'm going to have to buy a print from you and have it mounted on the wall. It'll make for such a great conversation piece, haha.
EDIT: In fact, I know just where I'll put in my office. Students will come in and ask about it and I'll say "Well, let me tell you a story..."
Glad to help. As I learned with my MarkovMTG Twitter account, there can be a lot of burnout on an idea like this at the beginning; however, most people appreciate a slow, once-a-day feed they can get from Twitter providing regular amusement for months to come, as something to chuckle as they pass it in the day's tweets. There's some good ones queued up!
I'd like to suggest Carnor Sprith; Slethward Bestroh; Light of the Bild; Strong Shield; Nevinyr Remain; Grirerer Knishing; or maybe Shring the Artist or Gravimite or Shisyfllen-Rand Thaters or Moltyr Defender or Prochonomance.
...OK, that's a lot. But those are the ones I would vote for, anyway.
This is so entertaining and interesting... and frightening. My late father was a published Professor of Computer and Information Sciences and this would have fascinated him. Well done.
The two best so far are definitely Slidshocking Krow and Carnor Sprith. I love the idea of inlect as a major mechanic in a block. So many deathtouch wolves.
And the render of Slidshocking Krow made me burst out laughing uncontrollably all over again. Perfect reminder text on Tromple.
Whatever you come up with, I'm going to have to buy a print from you and have it mounted on the wall. It'll make for such a great conversation piece, haha.
EDIT: In fact, I know just where I'll put in my office. Students will come in and ask about it and I'll say "Well, let me tell you a story..."
I will be sure to make whatever I end up painting into a print! It would be awesome if someone takes the art and puts it into a card like some of the other images in this thread!
I'd like to suggest Carnor Sprith; Slethward Bestroh; Light of the Bild; Strong Shield; Nevinyr Remain; Grirerer Knishing; or maybe Shring the Artist or Gravimite or Shisyfllen-Rand Thaters or Moltyr Defender or Prochonomance.
...OK, that's a lot. But those are the ones I would vote for, anyway.
I need a ruling on Skengi Hellldadietsn. Can you use him to generate infinite mana on the turn he enters play, or does the fifth copy of his ability to come off the stack fizzle because he is now in the graveyard?
I wonder, can someone do a series where its just fed the vanilla and french vanilla creatures? The p/t on almost all of these cards seem very appropriate with the mana costs. I wonder if it will pick up on how creature keywords are used. If so, this could be a very handy tool for filling out simple common creatures for a custom set.
Hey, wow. This is brilliant. It's also very similar to my (current) MsC dissertation.
I'm doing grammar induction on the M:tG corpus. In other words- I'm attempting to learn a grammar for Ability Text from cards.
I already have a hand-crafted grammar that can parse the AT for a handful of abilities (burn, bounce, exile, destroy etc). It's part of an AT interpreter I created for my degree dissertation. It parses Oracle text and executes it as scripts in my M:tG rule engine.
The network, effectively speaking, is performing grammar induction. In fact, I'm using neural networks precisely because I want to avoid having to hand-craft a grammar, such as was done here.
Since it's a grammar it also generates new AT strings, like your network does. In my case the generated AT tends to make a bit more sense, I guess because of the limited scope of the grammar. On the other hand, it can only generate one of the handful of abilities it "knows" about, so it has much less coverage than the AT generated by your network.
The beauty of using a learned system is that you can extend what it does simply by feeding it new inputs. It's very flexible.
Of course hand-crafting a grammar for AT is a huge pain, hence the new project, to learn the grammar automagickally, if I may be so bold
Excellent idea, and I share your sentiment completely. Now, I'll warn you, the problem with even sophisticated machine learning techniques is interpretability. That is, the implicit grammar that the network learns won't be clean or easy to understand if you try to reverse engineer an explicit grammar from it. Think of every weight in the network as its own micro-rule. It's a ***** to disentangle any meaning from it.
But then again, when it comes to neural networks, this should be expected. Similarly, if the human brain were so simple that we could easily describe it, we'd be so simple that we could not.
And yes, I'll shoot you an e-mail so you have my address if you want to speak further on the matter.
That's a problem with grammar induction in general it seems, it's possible to learn a grammar from text (with caveats) but the grammar you learn is usually very different to what linguists would expect for the same language.
That's because there's many different grammars that can describe a given language. I guess if you leave a learner running for ever eventually it would learn one that makes sense to you and me. But, well
Well, the learner will converge on a grammar that makes the most sense to the learner, and the learner may think differently than you and I.
I guess I have a plan on how to deal with this, but I'm going down the inductive logic programming path (and possibly the statistical relational learning also). The ANN approach might get there faster though, I don't know.
Magic English is a formal language, and as such one would assume that it is amenable to formal methods. And many formalists would consider it heresy to use statistical techniques for a formal language like we do informal, natural ones. After all, if you have absolute certainty about the meaning, why drag everything through the mud with fuzzy, statistical techniques?
I'll warn you, however, that I am a known heretic.
I will be sure to make whatever I end up painting into a print! It would be awesome if someone takes the art and puts it into a card like some of the other images in this thread!
I definitely have to vote for Slidshocking Krow, with the idea already posted in the thread of a goofy-looking dragon with parts rearranged or something similar.
I'd also request something in proportions to match an MtG card but that also works when cropped to the upper section - this way people could use it to generate both normal cards and full-art promo versions to use as prints or whatever.
Here's a thought in general - can anybody come up with a good set symbol for a theoretical Un-set entirely or mostly put together from computer-generated cards?
I will be sure to make whatever I end up painting into a print! It would be awesome if someone takes the art and puts it into a card like some of the other images in this thread!
I definitely have to vote for Slidshocking Krow, with the idea already posted in the thread of a goofy-looking dragon with parts rearranged or something similar.
I'd also request something in proportions to match an MtG card but that also works when cropped to the upper section - this way people could use it to generate both normal cards and full-art promo versions to use as prints or whatever.
Here's a thought in general - can anybody come up with a good set symbol for a theoretical Un-set entirely or mostly put together from computer-generated cards?
Go into the art assets for Binding of Isaac Rebirth and use the image for the item Missingno.
Private Mod Note
():
Rollback Post to RevisionRollBack
Rest in RIP My Signature, I guess. 2015-2016, we hardly knew ye.
Total side note but I like how people are discussing running this over itself.
Well, it has to be said that Kolaghan is clearly better than Kolaghan. Considering what BR decks usually want and how stiff the competition in this guild is, there is no way that I would run Kolaghan over Kolaghan! Kolaghan might make it in as the sixth guild card or so. Kolaghan on the other hand ranks several places below that.
It was so funny to me when they described this as a downgrade to the original Zurgo during the Pax East panel. I was thinking if this is a downgrade, they should really "downgrade" all legendary creatures. Haha.
My deck designing is quite concise at this point:
1. Come up with deck idea
2. Realize this idea is somehow fundamentally similar to another deck I have or that is commonly played in my group
3. Decide I don't want to disassemble one of my existing decks
4. Give up and do nothing
I don't see the point of this new shroud mechanic. It's strictly worse than Hexproof. Threshold is pretty bad too, Delirium is a much better mechanic and probably easier to activate.
Otherwise this card is a pretty neat guy. Dodges removal and grows into a Primal Huntbeast. 3/5
Ok, I have my RNN up and running (thanks to the excellent project documentation, and the very nicely formatted input provided by Talcos.) It's spitting out a checkpoint every hour, and has been running for three or so hours. I have several examples of cards from each checkpoint if anyone would be interested in seeing them.
Ok, I have my RNN up and running (thanks to the excellent project documentation, and the very nicely formatted input provided by Talcos.) It's spitting out a checkpoint every hour, and has been running for three or so hours. I have several examples of cards from each checkpoint if anyone would be interested in seeing them.
I think the answer is "very yes".
I should (hopefully) have some cards of my own to share shortly.
Private Mod Note
():
Rollback Post to RevisionRollBack
My Moderator Helpdesk
Currently Playing:
Legacy: Something U/W Controlish EDH Cube
Hypercube! A New EDH Deck Every Week(ish)!
Ok, I have my RNN up and running (thanks to the excellent project documentation, and the very nicely formatted input provided by Talcos.) It's spitting out a checkpoint every hour, and has been running for three or so hours. I have several examples of cards from each checkpoint if anyone would be interested in seeing them.
Awesome! I would love to see the results that you got.
Private Mod Note
():
Rollback Post to RevisionRollBack
I'm currently working on a small set called From the Catacombs.
Check out my new blog about artificial intelligence. Thanks!
To post a comment, please login or register a new account.
EDIT: Also, Izzy, very excited to see what you come up with. My vote's on Shring the Artist or Slidshocking Krow. But maybe it's not fair for me to vote, since they're the ones created by my own creation, haha.
My LinkedIn profile... thing (I have one of those now!).
My research team's webpage.
The mtg-rnn repo and the mtg-encode repo.
Izzy, this one plz from the original post:
Whatever you come up with, I'm going to have to buy a print from you and have it mounted on the wall. It'll make for such a great conversation piece, haha.
EDIT: In fact, I know just where I'll put in my office. Students will come in and ask about it and I'll say "Well, let me tell you a story..."
My LinkedIn profile... thing (I have one of those now!).
My research team's webpage.
The mtg-rnn repo and the mtg-encode repo.
...OK, that's a lot. But those are the ones I would vote for, anyway.
And the render of Slidshocking Krow made me burst out laughing uncontrollably all over again. Perfect reminder text on Tromple.
Yes. Very yes.
-
View User Profile
-
View Posts
-
Send Message
Official MTG ArtistI will be sure to make whatever I end up painting into a print!
-
View User Profile
-
View Posts
-
Send Message
Official MTG ArtistLOL, can you narrow it down to one, maybe?
Don't image leech.
The network, effectively speaking, is performing grammar induction. In fact, I'm using neural networks precisely because I want to avoid having to hand-craft a grammar, such as was done here.
The beauty of using a learned system is that you can extend what it does simply by feeding it new inputs. It's very flexible.
Excellent idea, and I share your sentiment completely. Now, I'll warn you, the problem with even sophisticated machine learning techniques is interpretability. That is, the implicit grammar that the network learns won't be clean or easy to understand if you try to reverse engineer an explicit grammar from it. Think of every weight in the network as its own micro-rule. It's a ***** to disentangle any meaning from it.
But then again, when it comes to neural networks, this should be expected. Similarly, if the human brain were so simple that we could easily describe it, we'd be so simple that we could not.
And yes, I'll shoot you an e-mail so you have my address if you want to speak further on the matter.
My LinkedIn profile... thing (I have one of those now!).
My research team's webpage.
The mtg-rnn repo and the mtg-encode repo.
Well, the learner will converge on a grammar that makes the most sense to the learner, and the learner may think differently than you and I.
Magic English is a formal language, and as such one would assume that it is amenable to formal methods. And many formalists would consider it heresy to use statistical techniques for a formal language like we do informal, natural ones. After all, if you have absolute certainty about the meaning, why drag everything through the mud with fuzzy, statistical techniques?
I'll warn you, however, that I am a known heretic.
Absolutely!
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 is so wrong.
He has none of the typical feline features.
This is the best card in the history of everything. Also the worst.
I'd also request something in proportions to match an MtG card but that also works when cropped to the upper section - this way people could use it to generate both normal cards and full-art promo versions to use as prints or whatever.
Here's a thought in general - can anybody come up with a good set symbol for a theoretical Un-set entirely or mostly put together from computer-generated cards?
Go into the art assets for Binding of Isaac Rebirth and use the image for the item Missingno.
-
View User Profile
-
View Posts
-
Send Message
StaffI think the answer is "very yes".
I should (hopefully) have some cards of my own to share shortly.
Currently Playing:
Legacy: Something U/W Controlish
EDH Cube
Hypercube! A New EDH Deck Every Week(ish)!
Awesome! I would love to see the results that you got.
Check out my new blog about artificial intelligence. Thanks!