A Japanese programmer has released a playable video game built from AI-generated art.
The programmer, who goes by the name Nao_u, created the 2D game over a three-day period. The result is a horizontal-scrolling sci-fi shooter called Shoon, which is similar to Nintendo classics such as Darius or R-Type. You can play the short game on a web browser.
But rather than hire a human artist to create the spaceships and vehicles in Shoon, Nao_u tapped an AI program called Midjourney to develop the in-game art, which includes the scrolling background and the title screen.
Similar to the DALL-E 2 program from the OpenAI Institute, Midjourney creates images from a text prompt. For example, if you type in “spooky Victorian haunted house,” Midjourney will attempt to draw an image depicting just that with sometimes stunning results.
Nao_u decided to use Midjourney to create various sci-fi battleships and other art assets for his Shoon game. “It's a matter of luck if you can get the material you want, but if you're good at it, the AI will create an inexhaustible amount of variations,” Nao_u wrote in a Twitter thread documenting the game’s development.
メカ系の素材は「starwars x-wing spacefighter flying from left to right in black space, armored core for answer, high quality」の呪文でガチャをひたすら回してたら偶然出た「背景が真っ黒で横向きのメカ」の絵がすべての祖先になってる (赤丸のやつ)
これを運に頼らず安定して再現したいなぁ… pic.twitter.com/xhyqjbZleA— Nao_u (@Nao_u_) August 14, 2022
The game demonstrates that it's possible AI algorithms can help video game makers complete some of the art for their projects, according to Nao_u. But Midjourney still has some serious limitations. For example, the program struggles to create images seen from a top-down angle, preventing Nao_u from creating a vertical shooting game.
“Another problem is that I can't make animations,” Nao_u added. “I'd like to make a side-view Contra-like run-and-gun game as well, but character grounding and animation are big hurdles. The advantage of the horizontal shooter is that you can cut out and paste the picture without animation.”
However, Nao_u is hopeful AI programs will advance to the point that they can create and animate 3D models: "There may come a day when unimportant assets are left to AI."