For a limited time, Google will let you create your own Chrome Dino game with GenDino. It’s a fun idea and an interesting example of how generative AI could be used for gaming.
GenDino is a Google Labs experiment that leverages the Imagen AI model. It’s identical to the standard Chrome Dino game, but it lets you choose your own character, enemies, and background elements. Just type what you want to see in the game, and GenDino will try to produce something that fits your prompt. You can also share the GenDino games that you generate.
There are some interesting guardrails here—your prompts cannot contain any spaces or special characters, and the names of some celebrities or companies are automatically rejected. However, I found that prompts like “Supermario” worked just fine.
I also enjoyed cycling through random prompts with the “I’m Feeling Lucky” button. And, while this AI’s sprites are surprisingly good, I found that some results were quite bad. You can refresh the output by re-entering a prompt. I had to do this for “Supermario,” since the first few outputs looked like garbled versions of Nintendo’s beloved mascot.
According to a Google Labs writeup, GenDino was created using Imagen 2 and StyleDrop. While the Imagen 2 model is a straightforward image generator, StyleDrop is a “fine tuning technique” that can evaluate a reference image and spit out similar-looking results. Google Labs simply pointed StyleDrop at the original Chrome Dino game and said, “make sprites that look like this.” Naturally, Google warns that GenDino’s output may be “incorrect, offensive, or inappropriate.” I tried to push the AI into this territory to no avail.
Last week, GenDino was offered as an exclusive experience for attendees at the I/O 2024 keynote. This public release of GenDino will only be available through May 28th, so you should play it while you still can. Note that GenDino requires a Google login. Thankfully, it works on both desktop and mobile devices, and you don’t need to use the Chrome browser to play it.
While GenDino is a rudimentary example of image generation in video games, it’s still a solid proof of concept. Game developers could use AI to produce textures, models, or other assets that are required in video games. This technology could also be useful to modders who want to build HD texture packs or other assets for existing games.
Source: Google Labs via 9to5Google