No one was surprised when Google said its latest version of Gemini is capable of doing a lot more than its predecessor. Exactly to what extent, though, was surprising to me. I typically don’t find myself particularly excited about new AI model releases, but after playing with Gemini 3 a bit, I stood corrected.
Still, that excitement only came after hearing that my editor had recreated the classic Minesweeper game with a simple prompt in Gemini 3. My interest was piqued. And after successfully creating my first simple game in a one-shot prompt, my gamer brain started firing on all cylinders.
I’d already had some experience with vibecoding in Gemini and walked away satisfied with the results, but the project itself was practical and more of a proof of concept than genuine interest. Throwing gaming into the mix, especially old-school games from my childhood — that changes things.
I decided to dive into Gemini’s vibe coding powers, and I couldn’t have had a better time. I never expected gold, and that’s not what I received. But it was enough to satisfy my retro gaming desires. Best of all, everything it spat out was playable within a web browser. From a reworked Chip’s Challenge to an 8-bit Tomba! and a very basic, 3D concept of the first Silent Hill game, this might be the most fun I’ve ever had with AI — and I barely scratched the surface.
Below, I’ll go over what I did, including how easy it was to code and make edits just by chatting with Gemini.
Gemini 3 feels more capable and thorough
Having a conversation with Gemini 3 feels different. The chatbot seems more aware of context and is able to address certain issues you might have as follow-up questions in the first go. Previous versions of Gemini could also do this, but Gemini 3 simply seems to be more thorough with its responses. And that’s just in regard to having a regular conversation about any particular subject. For my experiments, Gemini went above and beyond at not only explaining the technical (above my head) stuff, but also breaking it down into easier-to-understand pieces when it was creating a game for me.
For instance, in the middle of my testing, I asked if it could package one of these games into an Android Package Kit file format, so I could sideload it to my phone. Gemini told me it couldn’t do that, but asked if I’d like some steps on how to do it, and I said yes. After it generated the game I asked for, it created a separate document with steps on how to convert the HTML file it provided me into an APK using a variety of tools, broken down into effort levels. I didn’t go further beyond this point; I still felt like this was a Gemini I hadn’t chatted with before.
The cost of nostalgia
If you’ve ever wanted to relive old video games of your past and have visited a specialty retro gaming store, an unsettling reality comes into view. Old games can get incredibly expensive. In my visits to several stores in search of original Silent Hill titles, I’ve found I’d need to drop somewhere near $500 just to buy two of them — and that’s if I were lucky enough to find them in stock. And if you don’t have the console to play them on, expect to pay even more, or be okay with the titles becoming part of a collection and nothing more.
Trying to grasp at the tendrils of a younger self’s past joy comes at a cost, and it’s going to be out of reach for many people. Here’s where AI might be able to help preserve these games — or at least the memory of them — as they become harder to find and more expensive as they age. There’s a reason Gemini and other LLMs are so good at remaking old games: They’re trained on vast quantities of text and code, almost certainly including the source code for these games.
For the purposes of my tests, I intentionally asked Gemini to create “demakes” of old games: stripped-down, simplified versions that are essentially playable proof of concepts.
Gemini channeled my childhood with several ‘demakes’
Once I saw exactly what Gemini could do, the game was on. I immediately tapped into the old-school games that came preinstalled on Windows. For the most part, the end result with just one pass was short and playable, but the AI needed more instructions to make something worth anyone’s time. Nonetheless, I was impressed with many of Gemini’s first drafts and knew they’d only get better with refinement.
Chip’s Challenge
Gemini captured the essence of Chip’s Challenge in its old-school demake on its first try.
One game I haven’t seen get any sort of remake treatment in the App Store or Google Play is Chip’s Challenge. If you were a Windows 3.1 user, you probably know this game — even if you don’t think you do.
Run around to get keys to the next level, avoid enemies, the end. Gemini slapped together something reminiscent of this game in its first try, and I’m fairly confident it’d be able to give it a full remake with enough instructions.
Jezzball
Although there were some missing elements, Gemini 3’s first try gave me a working Jezzball game, and that’s all I asked for.
Jezzball is the perfect time killer and is as fun today as it was decades ago. Try to capture at least 75% of the space by building walls with lines without getting hit by one of the constantly bouncing balls within the space.
8-Bit Tomba!
Gemini’s first try of Tomba! made me realize that the sky is nearly the limit for my retro video game dreams.
One of my all-time favorite PlayStation 1 games is Tomba! I’d first played the first level on a demo collection disc for the first PlayStation and got hooked. I’ve emulated Tomba! on my phone and tablet over the years, but in 2024, a remake — officially called the Special Edition — was released and is now available on PC, PS5 and Switch. In Tomba!, you play as the titular pink-haired hero on a quest to save the world from evil pigs.
Gemini channeled the essence of what an 8-bit version of Tomba! would be on its first try. It was very stripped down, but the level name, Tomba’s abilities and small collectibles were all there.
Silent Hill
As a big Silent Hill fan, I was impressed both by how Gemini approached the creation of this game and the overall output.
Gemini made me so many fun and simple demakes, so I wanted to up the ante. I asked it to remake the first Silent Hill — in essence. I wasn’t expecting a full-fledged game here, but I wanted it to be something better than the previous games it had made, and in 3D this time.
Gemini understood the assignment, telling me: “To do Silent Hill justice, we cannot just make a game; we must create an atmosphere.”
Gemini created the game in three iterations and seemed to break it into stages. First, the atmosphere. The demake channels the original by adding a layer of fog to purposely create a barrier from what you can see ahead of you. What was a technical limitation from the original Silent Hill still makes for a great effect today. Add in the falling snow, and some structures resembling buildings and trees, and we were off to a good start for a demake of this nature.
Next came more important aspects of the Silent Hill series. The static-filled radio, the ability to swing a steel pipe and an enemy to fight. Gemini already had this in its queue, but wanted to check in with me to make sure that this was what I wanted — and I couldn’t have agreed more. The only part of this that needed adjusting was the enemy. In the first run through, it was just a cube-shaped blob, which hardly resembled anything intimidating enough to swing a pipe at. Gemini tweaked it to look a little bit more like a creature, even though it still resembled something far less scary than anything in the original games.
Then came the final touches. Silent Hill wouldn’t be the series it is without its music, and Gemini knew this. It created two audio files — one low frequency it dubbed “the drone” and one high frequency it dubbed “the scrape.” The music is unsettling enough to keep you slightly on edge. Another quintessential piece of the series is its puzzles, but Gemini didn’t go too complex here and just threw in a key that could be found while walking.
Finally, the shift to the “otherworld.” Gemini made this a manual option that shifts the fog to red and rusts the walls — all with a wailing siren in the background.
Here’s how to try this for yourself (and why you should)
If you want to ask Gemini to recreate an old video game — or any type of game (I had it make a touch version of Hungry Hungry Hippos) — you’ll want to make sure your settings are configured correctly.
Next to the text field where you talk to Gemini, you’ll want to look at the model you’re using. You can choose between “Fast” and “Thinking” models, but for creating games, you’ll probably want to choose the latter. The “Thinking” model (Gemini 3) is going to provide more in-depth information and will dig deeper into creating whatever you want it to. The Fast model, Gemini 2.5 Flash, won’t spend quite as long on the task.
When asking Gemini to create a game that’s playable on your phone or within a web browser, you’ll probably want to ask the AI to give you an HTML file.
I tried ChatGPT, too
In an incredibly unfair test, I asked ChatGPT (the free version) to recreate Tomba! In 8-bit, it quickly delivered a crude but playable, simple game. It lacked all the styling of the game that Gemini nailed on the first try, along with the ability to use his weapon, and more.
When I asked ChatGPT to make the level and characters look better, instead of updating the code, it generated an image. After I corrected it with my intentions, it went to work updating the style and colors of the game. This version was better in terms of style but lacked any sort of polish Gemini added when I asked the same exact question for the second iteration. I’d be curious to see how much better the latest ChatGPT models would fare in this test.
