Thank you for taking the time to play Hot Clock (2026).
Even without the use of AI, this game took hours and hours spread over several weeks and months in order to develop. I appreciate your curiosity and concern. The use of generative AI is a very very touchy subject in modern discourse. Rather than shy away from disclosing this stuff or shunning an emerging technology, I wanted to take the time to discuss with you exactly what I'm using AI for in regards to this game's development. My goal with disclosures like these is to allow people to make informed and fact-driven decisions on the media they engage with in respect to their personal philosophies.
I want to be clear: this is not some ironclad defense of AI or a starry-eyed sermon espousing how AI will solve all of our problems. Nor, as you might imagine, is it meant to be some kind of condemnation of the use of AI or a eulogy for "old-fashioned" game making. This is a statement confirming that I've used AI, and a detailed explanation of what I've used AI for, as well as what I think it is good and bad at in the context of making a game.
I believe that generative AI, like all technologies humanity has created, has its place. I believe we are starting to finally grasp the niches where AI can not only settle in appropriately, but be used greatly improve the lives of many around the world as opposed to just a chosen few. I also believe that AI has settled into some wildly inappropriate places and is enabling some very shrewd laymen to intrude into others' spaces as cynically and disrespectfully as possible. My goal is to find the right places to use this tool, because at the end of the day, AI is a pretty versatile tool that can open new doors for those who can use it properly and responsibly.
I'm going to break this paper down into some fundamental "chunks" of gamedev: Game, Scenario, Art, Audio, and Program. Not all games will even approach the use of AI in the same way as mine. So just because some aspect of AI works for me, doesn't mean it can or should work for you, the reader.
All AI generation was provided by ChatGPT models 4o, 5, and 5.5. No, I'm not interested in trying Claude right now. I don't care if it's supposed to be better at coding.
Hot Clock is a very simple premise: navigate difficult obstacles through trial and error leveraging timing and memorization by using a forgiving respawn mechanic. It's Super Meat Boy.
Because this game was technically made a decade before AI, there isn't really anything to disclose here. Pretty much all of the level design and geometry got ported as-is.
Nothing about this game concept or the core loop was created with generative AI.
Hot Clock does not have a story ;) Because of this, no generative AI was used for story purposes.
And even if it did have a story, which it doesn't ;) then I wouldn't be interested in using generative AI to make it.
No generative AI was used in the production of Hot Clock's music or sounds.
Everything was composed by myself in Acoustica Mixcraft Pro Studio 9 between 2022 and 2025. Audacity is also used for minor sound editing.
The lo-fi sample rate of the music came about some time ago, inspired by the album Space Hymns, which originally released in 24kbps MP3 format. This absurdly low quality gave the songs very limited acoustic headroom, but when combined with stripped-back arrangements and a good tonal balance between stoic nostalgia and unsettling ambient noise, creates a captivating album that is greater than the sum of its simple parts.
Since I discovered Brother Android in the late 2000s, I've had a small interest in "wringing" songs in such a fashion, mixing them down in as low a quality as the DAW would let me and seeing what came out on the other side. My execution is mostly a callback to old Flash games, which made extensive use of heavily-compressed, legally-dubious music loops. I found the constraints to be surprisingly adaptable, although I lost a lot of low-end headroom in my sound mixes. Also, GameMaker gets really fussy when you try to convert audio samples below its means.
The best part, though, is that the limitations of the lo-fi style allowed me to save room for certain...surprises.
Hot Clock's audio is 99% sourced through Freesound, and the jump sound effect is sourced through someone I found on Gumroad or something. Because this was 13 years ago, I don't have any of the receipts anymore. Oh darn.
Hot Clock uses generative AI for minor asset cleanup and placeholder asset generation; everything else is hand-drawn or hand-sourced.
Art is a very frustrating endeavor for me. Because of that, I took a hamfisted approach when remastering this game. What this means is that I made some deliberate decisions, such as parallax effects, increased readability, and themed worlds. The rest of the decisions, however, were at the mercy of the tides.
Why are there realistic circuit towers everywhere in World 3? What's that stuff in the background of World 2? Why did you put spoons in that one level? Why does the player character walk like a Mooninite? I don't know.
To break things down a little, let's talk about what I sourced. The assets drawn by me include the player character, obstacle sprites, and tilesets. Text fonts are sourced through Dafont. Anything that even remotely looks like an image is sourced through Pixabay. Most art assets were modified in some way, shape or form with paint.NET in order to achieve a desired look.
Here are a few sprites as they appear in the 2013 version of the game: (1) (2) (3) (4)
And here are those same sprites remade for the 2026 version of the game: (1) (2) (3) (4)
That spring sprite is my masterpiece. I drew and animated it myself and I will hear nothing but your finest constructive criticism.
With all of that said, I would like to disclose the singular asset in the game that uses generative AI. This was more experimental than anything, and I chose to keep it because I believe it violates no significant moral or ethical boundaries: the humble spoon. This spoon was sourced from a Pixabay image, and I was curious to see if ChatGPT could crop it out. I suppose it did so valiantly, even if "cropping it out" may have just meant "redraw it without everything around it." I have a lot of experience doing basic cutouts, but this was an example of having someone else do the legwork for a change. I chalk its continued use up to the fact that the game's worlds are a little inconsistent in visual theming beyond nebulous "vibes." I definitely would not have used AI for this type of thing in a commercial or professional title.
With that out of the way, let's talk about placeholders. While in development, Hot Clock did use three AI-generated assets while their final assets were awaiting manual production. These being the level end medals, the ending screen, and the forbidden object.
The level end medals were generated near the end of development, when I elected for a level end sequence that was a little more substantial and rewarding for the player than the sequence present in the original game. What's interesting to note is that ChatGPT can generate assets on a transparent background. That said, I asked for the sprites to be 32x32 and none of them were even close. Also, the platinum medal looks kind of like a sorry silver. By the time the art pass was ready, I moved on to creating star-shaped medals using a paint.NET primitive and some basic GML image manipulation.
The ending scene was a mess. The asset I generated was a splash screen that used Japanese Kanji. The original idea of the ending scene was to reference Neon Genesis Evangelion, but I got completely lost in the details, and by the time I had gotten anywhere with it, my brain sorta woke up and asked what I was doing, and I didn't have a good answer, so I scrapped it and made something else by myself. I think the biggest factor was that I don't know any Japanese, so attempting to have it in the game with zero knowledge of what it read felt a little ignorant, in my opinion. By all means, if you or someone you love is Japanese, have them read this graphic and offer their brutally honest feedback.
What I will say is that the way ChatGPT walked me through its process was nothing short of fascinating. It broke down what each character in the text means in both its literal and cultural context. Japanese is a language with a single culture tied to it, and seeing how the words were explained felt like I was learning a bit of Japanese social customs on top of some basic language. This, of course, comes with the caveat that this is AI talking, and it's entirely possible that none of what it explained to me is true.
There is one final placeholder asset that was generated with AI. This rotating sphere was generated using the now-defunct Sora video generator. It was used for a special object and was actually intended to be in the final release, but near the end of development I began having serious objections to the inclusion of AI-generated art assets in the final build, so I did some photoshopping of a Lego picture and made something distinct enough. Most of the animating in this case was done with basic GML sprite manipulation.
Also, here's the old tileset vs the new tileset. I hand-drew it instead of doing the simple color palette as a way to challenge myself.
Hot Clock makes sparse use of AI-generated code. AI was also used for diagnoses of various bugs and glitches. No vibe-coding was used in the development of Hot Clock.
I've been coding for quite some time. I have a particular way of doing things and I'm a control freak. Because of this, I get a little obsessive when adding outside code to my projects. Usually that means refactoring everything to fit my coding style and existing infrastructure. Outside of a few key exceptions such a HTML5 fullscreen support, I am a bit loath to use 100% external code without at least some kind of refinement process.
Hot clock uses very few AI-generated scripts, which I've posted here. These are mainly for edge cases in which the solutions were a little trickier to grasp. They also represent situations where I know I've written a function for something before, but because it's buried in some other project somewhere, I found it easier to just have an AI generate a very clean function I can drop in and just call it good.
One of AI's most useful cases for me was as an "interactive librarian." I could ask it about various GameMaker functions and it was really good at explaining them and giving me examples of their use. As comprehensive as the official manual is, I find it to often be a little wishy-washy with data types and the explanations aren't always clear.
I also used ChatGPT to help fix random issues I was having, such as lerp roundoff errors, explanations of ancient functions, or the tunnel effect in Level 4-5.
