Mark Andrade

Journal

Cloud Phone Games

About a month ago I ported several games to the Cloud Phone platform. The website can be found at https://cloudphone.taara.games.

There's some input latency and the FPS is low. (though the games are running at 60FPS on their server). It's an interesting platform. I might create some more games for this platform in the future.

Photo of the Nokia TA-1563 running Cloud Phone showing published and dev icons of my games
Photo of the Nokia TA-1563 running Cloud Phone showing published and dev icons of my games
Close up photo of the Nokia TA-1563 screen running Cloud Phone showing published and dev icons of my games
Close up photo of the Nokia TA-1563 screen running Cloud Phone showing published and dev icons of my games. Not many pixels on this thing.

The AI GameDev Adventure: A Messy First Step

I threw myself into this AI experiment with wild anticipation. I didn’t need AI to do all the heavy lifting—just enough to take the edge off. Claude Sonnet for code, DrawThings for art. Easy, right? Yeah, not so much. After a week of messing around, I realized this was going to be more of a wrestling match than a smooth collaboration.

Pixel art, in theory, should’ve been fun. And I’ll admit, there were moments when it was—just not exactly the way I expected. I asked DrawThings for a red serpent dragon in nice, neat 16x8 frame segments. Instead, what I got was… well, these intestine-looking monsters that look more like something dredged up from a biology textbook than a game sprites.

Nope!
Ummm Nope!!
Oh Hell No!!!!!

Each new attempt gave me a slightly less terrible version of the same.

I tweaked the prompts, cursed at the screen, and finally let it run free to generate whatever nonsense it felt like. “Maybe this time it’ll work,” I thought. After what felt like an eternity of frustration (and some serious self-questioning), I let DrawThings run wild, generating a bunch of images. Some were fun, most were tragic. At least I got a couple of renders that didn’t make me want to immediately give up. But let’s not get it twisted—“decent” here means the difference between a solidly passable turd and getting something that looks like the remnants of a bad Taco Tuesday.

I figured if I kept at it, one of these images would strike gold. But nope. The AI just didn’t seem to get what I had in mind.

LOL WTF?!
Got fed up and just asked for some dragons to see what would happen.
Composite of some random sprites from DrawThings that I kept around.

Eventually, I gave up on coaxing it to make something useful and shifted gears to manual mode. A couple of tweaks in Affinity Designer I had art I could actually work with.

And it’s all I needed.

Zoomed in look at the two sprite frames of the dragon

Sure, it took longer than I wanted, but at least I didn’t feel like I was banging my head against the wall. Sometimes, you’ve just got to accept that making the art yourself is quicker than waiting for AI to catch up. Plus, there’s that strange temptation to keep pushing the “generate” button like you’re gambling for the perfect sprite. Spoiler: that strategy didn’t pay off.

Actual artboard size and some sprites I'm currently working with

But man, it’s gonna be a ride. The next step? I’ll keep grinding, watching YouTube tutorials like I’m cramming for some bizarre AI exam. Maybe one day I’ll get good enough at prompting to make something that works.

Next I start the boiler plate code since the background art’s looking half-decent.