Rowan, Scholar of Sparks | Strixhaven Schools of Mages
16, Jun, 23

MTG Designer Reveals Future MTG Sets Could Be Made with AI!

Share on facebook
Share on twitter
Share on whatsapp
The machines are taking over!
Article at a Glance

While we may receive upwards of six brand new MTG sets each yet, designing Magic: the Gathering takes a whole lot of work. Typically, Wizards of the Coast has a two-and-a-half-year development cycle, with rigorous design and playtesting. While this long development is vital to ensure sets aren’t riddled with game-breaking combos, it’s understandable Wizards may want to expedite the process. 

Previously, when looking to speed up development, throwing money or personnel at the problem was the best bet. Recently, however, Artificial Intelligence (AI) has been making waves by potentially accelerating development like never before. In just seconds, AI programs such as ChatGPT can complete work that would have otherwise taken hours, if not days. 

Due to this incredible speed, it’s no wonder that so many industries have jumped aboard the AI bandwagon. While this technically is undeniably impressive, it does have its limitations. Complex, interconnected, and innovation-driven games such as MTG, for instance, as a step too far for an AI to handle. 

While this is certainly true at the moment, according to MTG’s Lead Designer, it may only be a matter of time before AI is helping if not entirely developing, MTG sets.


D00-DL, Caricaturist | Unfinity
D00-DL, Caricaturist | Unfinity

Toward the start of the AI boom last year, we covered how programs such as DALL-E-2 and Midjourney could revolutionize art in MTG. Providing impressive generative artworks in seconds, AI programs such as Midjourney weren’t just effective, they were also cheap. In the ten months since then, AI has only gotten better and better at an almost alarming rate. 

Currently, while it’s far from the only program out there, ChatGPT by OpenAI is the go-to Artificial Intelligence program. Providing text, rather than images, ChatGPT has been used by coders, writers, and even lawyers with varying degrees of success. This program can even make what appear to be rather compelling MTG cards in an instant. 

To demonstrate this, we asked ChatGPT to try and predict the future of MTG for us. Working off the prompt “Create a Magic: the Gathering card for the character Rowan Kenrith for the set Wilds of Eldraine,” ChatGPT provided a new card in just a few seconds. Here’s what it came up with.

Rowan Kenrith, Ignition Mage

Mana Cost: 3RR

Type: Legendary Planeswalker – Rowan

Loyalty: 4

+1: Rowan Kenrith deals 1 damage to any target.

-2: Rowan Kenrith deals 2 damage to each creature without flying.

-7: You get an emblem with “Whenever you cast an instant or sorcery spell, Rowan Kenrith deals 5 damage to any target.”

Flavor Text: “Fire is a language I understand. With every spark, I ignite the world.”

As we’ll talk about shortly, this AI-generated MTG card is far from perfect. That being said, for a few seconds of work, it’s nevertheless an impressive feat. It even came up with flavor text, after all, and the theme feels rather apt. Due to the speed of ChatGPT, it seems completely understandable that Wizards may want to make use of this software. According to MTG’s Lead Designer, Mark Rosewater, that seems to be exactly the plan.

Responding to a recent question on their Q&A blog, Blogatog, Mark Rosewater revealed their current stance on AI. Stating, “I believe there will come a day where AI is a tool we use for design,” it appears that it’s only a matter of time before the first AI-designed MTG card is in Magic. That being said, however, before then, Rosewater also noted that “there are a whole host of issues to work through before we get there.”

Dubious Designs

Klothys’s Design | Theros Beyond Death
Klothys’s Design | Theros Beyond Death

To highlight some of those issues, we can look back to our fictitious Rowan Kenrith, Ignition Mage. On the surface, this card draws from the character’s lore and past abilities to create something new and interesting. In the grand context of MTG, however, this card is totally out of place. 

Following on from the events of March of the Machine: The Aftermath, Planeswalkers in MTG have all but disappeared. There are still some characters who have retained their spark, such as Ajani, however, most Planeswalkers have been desparked. While we don’t know for certain just yet, this is likely the case for both Rowan and Will Kenrith. If this theory is correct, then Rowan Kenrith, Ignition Mage is fundamentally flawed.

For better or worse, AI can be used to fix this problem, as all it takes is another prompt. Asking ChatGPT to “Create another version of Rowan Kenrith which is not a Planeswalker card” it quickly provided an alternative. Once again, we’ll put the card ChatGPT made below. 

Rowan Kenrith, Enchantress of Flames

Mana Cost: 2RR

Type: Legendary Creature – Human Wizard

Power/Toughness: 3/3

Whenever you cast an instant or sorcery spell, Rowan Kenrith deals 1 damage to any target.

{R}{R}{R}: Rowan Kenrith gets +1/+0 until end of turn.

Flavor Text: “The flicker of a spark, the blaze of a conflagration. All it takes is a single spark to ignite the inferno.”

While they’re not a Planeswalker this time, the AI-made Rowan Kenrith, Enchantress of Flames isn’t exactly good either. Featuring the abilities of Rockslide Sorcerer except with terribly expensive firebreathing, Rowan Kenrith, Enchantress of Flames just isn’t a good MTG card. Somewhat bizarrely, ChatGPT also made this care a rare, which would definitely be odd considering Rowan Kenrith’s expected prominence. 

From these mistakes, it’s clear to see that ChatGPT’s current MTG card-creating abilities aren’t quite perfect. Once again, however, the flaws can be mitigated by additional prompts and generation. For instance, to make this card more viable, we simply asked ChatGPT to “Make this card more powerful.” As always, the results of this generation are below. 

Rowan Kenrith, Flameweaver Sovereign

Mana Cost: 1RRR

Type: Legendary Creature – Human Wizard

Power/Toughness: 4/4

Whenever you cast an instant or sorcery spell, Rowan Kenrith deals 2 damage to any target.

{R}{R}{R}: Rowan Kenrith gets +2/+0 and gains trample until end of turn.

{2}{R}{R}: Rowan Kenrith deals damage to any target equal to the number of instant and sorcery spells you’ve cast this turn.

Flavor Text: “From flickering ember to roaring inferno, my flames shall consume all who oppose me.”

Mechanical Malfunction

Malfunction | Kaladesh
Malfunction | Kaladesh

While Rowan Kenrith, Flameweaver Sovereign may not be perfect, these generations combined took less than a minute. Compare this to the hours that it’d take to design an MTG card traditionally, and ChatGPT clearly has an advantage. Even if a card isn’t perfect, running another dozen of generations will still only take a fraction of the usual time. 

Due to this distinct speed advantage, it may seem like ChatGPT may be the way to go. In reality, however, this program, while impressive, isn’t really doing anything new. Rather than having its own creativity, ChatGPT instead remixes what it knows into something new. Considering it has the internet at its disposal, ChatGPT certainly has a lot of information to pull from. This, however, is still not the same as raw human creativity. 

To demonstrate this, we can simply ask ChatGPT to “Create a brand new mechanic for the Magic: the Gathering set, Wilds of Eldraine.” Responding to this prompt, the first mechanic that ChatGPU came up with was “Mystical Enchantment.” Activated as an enter-the-battlefield effect, this mechanic had six different sub-effects which could trigger. 

As you can see below, nothing about this mechanic, from the die rolling to the sub-abilities, is inherently new. Instead, ChatGPT has simply taken a bunch of preexisting concepts and fused them together. 

Mystical Enchantment (Whenever this creature enters the battlefield or you cast a spell with Mystical Enchantment, roll a six-sided die or use a random number generator. Choose one of the following enchantment subtypes at random and apply its effect to [Cardname]):

  • Enchantment of Vigor: [Cardname] gains +1/+1 until end of turn.
  • Enchantment of Illusion: [Cardname] gains flying until end of turn.
  • Enchantment of Resilience: [Cardname] gains hexproof until end of turn.
  • Enchantment of Charisma: Whenever [Cardname] deals combat damage to a player this turn, draw a card.
  • Enchantment of Cunning: Tap target creature an opponent controls. It doesn’t untap during its controller’s next untap step.
  • Enchantment of Transformation: Transform [Cardname] into an enchantment land with “{T}: Add one mana of any color to your mana pool.”

To see what more it could do, we kept asking ChatGPT to keep coming up with new mechanics a few more times. On the third attempt, the AI program churned out a new mechanic called “Arcane Revelry.” As you can see below, this supposedly brand-new mechanic is basically just Storm which was first seen in 2003.

Arcane Revelry (When you cast this spell, if it’s not the first spell you’ve cast this turn, copy it for each other spell you’ve cast this turn. You may choose new targets for each copy.)” 

The Future Isn’t Here Yet

Sword of Once and Future | March of the Machine
Sword of Once and Future | March of the Machine

Just like the cards we had ChatGPT create, it’s safe to say these mechanics aren’t perfect. This, however, doesn’t mean that AI doesn’t have a place in the development of MTG. After all, due to the speed of ChatGPT, it’s possible the software could simply brute force creativity. By asking the program to develop a hundred of a thousand new mechanics, chances are one of them would be viable. 

While this approach could certainly result in some interesting results, ChatGPT doesn’t have to do absolutely everything. Instead, AI programs could be used to simply assist in the development of MTG sets. Whether by generating boring ol’ common cards or by expediting the playtesting process, AI could be an incredibly useful tool.

Since creativity and innovation are at the heart of MTG, this seems like the most likely use of AI at Wizards. Rather than replacing designers outright, AI would instead merely assist where it made sense. Even before then, however, AI programs will have to get a lot better. Due to the current rate of their development, it may not be long before that happens. For the time being, however, the jobs of MTG designers are safe and sound. 

Read More: Nostalgic Graveyard Combo Deck Reemerges in Massive Event!

*MTG Rocks is supported by its audience. When you purchase through links on our site, we may earn an affiliate commission. Learn more