Retro Games

AI Art vs Human Art – A Super Mario Bros Experiment

AI Art vs Human Art – A Super Mario Bros Experiment

One of the hottest topics in the world of gaming right now is AI art. Technology has come a long way in recent years; but can it truly be used to build a new game? How does AI art compare to human art? Let’s explore AI Art vs Human Art!

An AI generated piece of art

What is AI Art?

What AI art does is attempt to interpret your brief and creates a piece of art that blends existing art into a new piece.

You could type in something like “Mario fighting Bowser in the future, cyberpunk, ghibi style” and in a few minutes it goes off and brews up a set of pieces with what it thinks you’re talking about.

We recently created futuristic AI Zelda art using Midjourney technology, where AI created this freaky dystopic world full of robots and Hylians. The video was met with a little backlash, with criticism that AI art is taking away from human art.

Zelda AI vs Human art

Whilst AI art creations are impressive, it doesn’t quite capture that human touch, the nuance and context. But at the same time, it’s always fun to see what AI can create. AI art is a roulette of interesting, weird, and downright wrong art pieces.

Creating a Futuristic Mario Game with AI

In a recent AI Art vs Human Art video we decided to try Midjourney tech once again to brew up a Mario world. We also tried to compare this to human-created concept art from past Mario games. Can AI art truly capture that heart, that unique art-style of human artists?

Princess Peach's castle made through AI

Personally, I don’t think so. Whilst AI art is impressive, it loses that signature and uniqueness that individual artists have.

But as an experiment and a discussion point, it’s interesting to see what AI can come up with, the limitations and that odd, almost soulless disconnect because it’s missing the human understanding.

AI Art vs Human Art Experiment – Architecture

The first AI art vs Human art experiment looks at architecture. it’s something AI does well as it’s usually a more concrete concept. Literally. We took a piece of concept art from Super Mario Galaxy and asked the AI to create its own interpretation. The results are close. The AI understood that it’s a castle, it’s floating and in space. But does it look like Mario or Nintendo in art direction? Not quite. That’s perhaps what’s missing from the AI’s understanding.

A look at AI art vs human art, with a castle

AI Art vs Human Art Experiment – The Toads

Let’s try those cheeky Toads next. Another piece of concept art from the Super Mario Galaxy series; it’s a simple one with a lot of charm and movement. Soft colours, distinctive in the game’s art direction. It’s a pastel, almost dreamy look. The AI attempt is impressive, it feels a bit like a Pixar design in style and it has hints of Mario with a shell shape, but those are not Toad characters!

The toads from the Mario series - AI art vs human art

AI Art vs Human Art Experiment – A Castle

The next thing to try in our AI art vs Human art experiment: Princess Peach’s castle. Ever since Super Mario 64, the castle has had such a distinctive look. Grass, a moat, grey and reds, flags, and that distinctive Princess Peach portrait in the middle. If you see it, you know you’re back in the Mushroom Kingdom.

Let’s see how AI catches that history and understanding. The result is striking, visually impressive. The AI understands castle, the moon, but again it misses what makes it Princess Peach’s castle. Without context it could be any castle that’s on a floating island. Conceptually it also glitches out with the moon having cut into the castle when the AI tried to stich everything together.

AI art vs human art with Princess Peach's castle

Where AI Goes Wrong

The biggest challenge with AI and where the human understanding truly comes into play is when you ask AI to create a specific scene. This next one is the perfect example of where the technology does fall short.

If you say to an artist, “paint Mario underwater, with fish”, it’s easy for a human artist to know what that’ll look like. The AI interpretation is, well, a bit different. AI understood the underwater scene, but it didn’t quite distinguish Mario swimming with fish.

Can AI art handle underwater scenes? A picture of AI art vs human art for underwater scenes

Can AI understand Multiple Characters?

The next experiment goes a little further with a more complex scene; let’s try to recreate a classic piece of art from Super Mario Bros. How do you describe this? There are so many elements to consider – Mario jumping, Bowser, the castle, the baddies. A difficult one to put into words!

The AI attempt comes close given it does understand the Mario brief, but what on earth is wrong with Mario’s body and proportions? The castle, baddies and Bowser are missing. It’s an interesting result, but it just doesn’t quite hit the brief.

 

Can AI understand Specific Actions?

Specific actions are always something that’s worth testing in AI. Let’s take a simple one, that classic Super Mario 64 scene where Mario grabs Bowser by the tail and throws him. The AI result is impressive, it understood the stone floor and the two characters, but again it’s just not quite right – that is not quite Bowser – it’s likely it’s stitched the two together from different art sources, so it doesn’t quite blend correctly. No attempt showed Mario holding Bowser’s tail because of this.

A picture of AI art vs human art, with more complex actions - a picture of Mario fighting Bowser

What about a simpler scene, but with – again – slightly more complex actions. This cheery scene with the two brothers playing Switch? The AI gave it a good go, but just didn’t quite get it right. Why is Mario in green? A nice-looking image, but it’s a weird mashup of concepts that just doesn’t quite blend  the words together correctly.

Two AI images - one is Mario and Lugi at the beach by human art and the other is AI art

AI Art vs Human Art Experiment – 3D Renders

Let’s venture into 3D renders now. Painted and drawn art is one area of AI, but one thing that the tech does is churn out rendered 3D art quickly. But can it do it right? I tried to recreate the upside-down Mario Kart 8 render through Midjourney, and it created this disturbing monstrosity. It’s definitely Mario but look at those eyes – and funnily enough there’s a tyre in his hat and parts of the sourced images floating in space. Again, it works but it didn’t get nuances in brief, like being upside down.

Two images of Mario Kart, one made by humans and rendered in 3D, the other by AI

Can AI Art replace Human Art?

The AI art looks great, but could it truly replace human art? I don’t think so. Not yet anyway. It’s an interesting experiment, but it proves that there’s still something missing. The human understanding in concepts, in creativity and heart.

There is a fear that AI could replace human art. Ethically there’s a big question mark given it sources and tries to blend existing human art, but I don’t think AI art is quite there yet in its understanding and creativity.

Be sure to watch our AI Art vs Human Art | A Mario Bros Experiment video.

What do you think of AI art – let us know with a comment!

mm

About J

J is TeeChu's founder and resident artist, a big Sonic and Zelda fan. Born and raised on hedgehogs and forest adventures!

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.