Time to do a recap of how much I do and don't use A.I. in the world of AGameAWeek.

-=-=-

Me and Michael Fernie were having an email chatter about AI and how much I'm making use of it in the world of AGameAWeek, so I thought I should post an update about what I am or aren't doing.

1. Coding - No

Honestly, I enjoy coding.
It's been the thing that's kept me sane through all of the horrific things that have happened to me over the past decade or so. It's my hobby, but it's also my sanity check, and after brain surgery sessions has been my self-diagnoses tool! (Can I still code a basic version of pong in my head, being something I've done a couple of times in hospital!)

Coding what I do each week, there's always a fun maths/logic puzzle that I have to solve to make each game work, and have the end result be fun to play. Trying to solve the elements that a game needs in order to function correctly is what I love to do.
Like solving a crossword, fitting together a jigsaw, or other puzzle things, this is how my head functions when coding.

So, no.
As far as what the A.I. CAN do, I don't tend to use the A.I.s to do coding for me.

On occasion, when making a little scripted tool for myself (like parsing data and transforming it from one form to another) I might lazily ask an A.I. to help me, but only because there's no "problem solving" there. It's something I've coded a hundred times before. There's no "need" to figure out a new way to do things. It's just to be used as a tool.
So on those occasions, I'll ask the A.I. to write a small function or something. That can be lazy.
.. And I'll often slap myself on the wrist for being so lazy. But these things are just tools, and I think tools are OK for the A.I. Right?!
ChatGPT is usually good enough for doing those types of coding functions.
But, as with all A.I. coders, don't take it at face value, always double check what it's given you, and for the love of all that is holy, do NOT paste any code it gives you directly into a live product!!!

Every few months, I'll give the majority of JSE's script to the various chatbots, and the one "optimisation" that they all seem to be convinced is a good idea, is to remove all the sanity/security checks from the script because they're slowing it down!!!
Um.. No!!!

2. Ideas - Yes!!

It's probably no surprise that trying to write AGameAWeek, and having done it for as long as I have, I kinda ran out of fresh ideas around about a decade or two ago!
Over the past year or so, I've started letting the A.I.s come up with a few ideas. This has been done in multiple ways.
Sometimes I'll have a friendly chat with an AI, discuss the sorts of games I make, then ask for suggestions for new games I might not have done. Another method is to use the Image generators.. Ask it for boxart or screenshots from long forgotten NES games, and you'll occasionally stumble upon a crazy idea, in there, somewhere.

Other times, I'll guide the chatbots into a certain direction by suggesting things like control schemes, or graphics styles, or gameplay elements. Things that constrain the sorts of ideas the A.I. can come up with. These can lead to obscure ideas, and I've also found that then feeding an idea from one A.I. chatbot to another, can end up with a nice three-way idea plan emerging.

In fact, the game I'm currently working on for Shoebox is one that started with an idea from DeepSeek, but that ChatGPT added a couple of elements to.
What DeepSeek thought might make for a fun action game, ChatGPT suggested it might work well as a Word game, because ChatGPT's memory knows the sort of games that tend to work well in Shoebox.
As each model grows, and I'm cramming them with information about the world of AGameAWeek, the ideas are becoming marginally more workable in terms of the sorts of games I've been able to code.
.. But that has its own issues. ChatGPT's now stuck on "Hey, this could make a nice wordgame!" ideas, and after a while, it's going to be as stuck in a rut as I am!!

ChatGPT's still my goto, though, mostly because it feels like it's more sensible than the other options! Sure, you can ask it specifically for surreal answers, but even then they can be more logic-driven, and based on its memory of you.

As far as experimenting with chatbots locally, there's a great piece of free software called PrivateLLM which I've got installed on the Mac and iPad. I can generally just about run the low-end chatbots on those, using my base-level Apple Silicon systems, but some of them can be quite good at coming up with off-the-wall ideas, as a result! Nothing better than a corrupt, struggling A.I. to hand out the oddities

3. Graphics - Kinda, Sometimes

The keyword "Spritesheet" is incredibly powerful, especially as the various art generator tools have been advancing.
It used to be that you'd get mushy "pretend" anti-alias pixels, and things looked somewhat faked, you can now get some amazingly high quality sprites out of the things.

Take Google FX ImageFX. (Free to use, assuming its available in your country)
If you ask it for "8-bit 4 colour Spritesheet : An anthropomorphic snare drum, platform game, running, jumping, rolling" you get...


8-bit 4 colour Spritesheet : An anthropomorphic snare drum, platform game, running, jumping, rolling

And, that's ... almost useable.
A year ago, the art generators would "try" to make pixelart and spritesheets, but would've generated something that would require a significant amount of reworking to, but nowadays things are much more recognisable, and the tools seem to understand colour limits and aliasing and such..

Of course, with the language I'm using (JSE) I still need to reduce the colours and fit the sprites down to 32x32 sprites.
I'll usually redraw things in my particular style, so they fit better within my sorts of games. There's been a few instances where I've grabbed sprites (say fishes!) directly from the output, but only when they're not really recognisable anyway!!

Mostly it's a case of "Ok, that sprite looks cute, I'll pinch that idea" and then I'll end up redrawing it based on the image, or the idea that the image pops into my head, rather than directly copying them straight into the game.
..
After all, my games do have a particular ..... um.. .. style.. !!!

4. Art - More and More

The daily blogs for the past couple of years have featured AI generated pictures of my "Greenie" (aka Derek) character.
Trying to get an AI to reproduce an animated character used to be really really hard.


Two years ago!

So I spent a good number of weeks playing with Stable Diffusion and LoRA.
DrawThings is a great free piece of software on Mac and iOS for using the various Stable Diffusion models, but to train the LoRA (which is the way of training a mini-model of your own people/items/characters) I needed to make use of a bunch of different Google Colab scripts, which I could only run about 2 or 3 times a day without running out of Free Functionality! (These seem to change and evolve every week or so, so you'll have to track down your own trainer!!)

I would draw a version of Greenie, feed it through a LoRA trainer, wait for about an hour at a time, then see how the character ended up being interpreted. I'd then try my best to "fix" what I'd drawn and do it all again!
I tried again, and again, and over and over, for what felt like months, in order to get the diffusion model working, and understandably and recognisably Greenie.

I also learned very early on that I couldn't call him Greenie, because that would turn the character Green, so I ended up giving him the name Derek!!!
Eventually, I learned how to draw Greenie "properly" from my experiments with Stable Diffusion, just as much as it learned from me what he oughta look like, and the Greenie's that I drew were slowly becoming more and more compatible with the A.I.


From there, I fed my versions of Greenie into ArtFlow.ai, which is about £100 a year, but which would come up with a beautifully stylised CGI/Claymation version of Greenie.
ArtFlow is meant to reproduce humans, but on a whim I decided to give it a go, and was gobsmacked at just how good it could draw Greenie.
.. As were the devs of ArtFlow, over on their Discord. They couldn't believe how well Greenie was being generated by their own software. "It's not meant for that... ... oh!"
I guess it's mostly because of the work I put into learning how to draw him "properly" for the AI's to understand him.
Nowadays when I generate him willy-nilly, it's easy to forget just how much work I put into making it so easy for myself.
Laying the groundwork is never a bad thing, and if it means I can now draw Greenie, myself, to a moderately better standard, then so be it!


I DREW DAT!

More recently I've been using Replicate (and the Flux model) to draw Greenies. The same inputs were given to the engine, but this time the result tends to be more in my style, rather than the 3D CGI looking Greenies that ArtFlow makes. Cost wise, Replicate is a "as you use it" setup, so you pay a monthly charge depending on how much you've used it. Since Flux is as blazingly fast as it is, the output's are coming out around 2 or 3p each, and I'm racking up about £5 or £6 per month as a result. Which isn't "too" bad.
Could be worse!!!
And the quality of the "proper" models is wonderful.

The daily blog posts have had an image every day, since, and they're generated using whichever AI looks better each morning. Usually Replicate, lately.

Evil!

Is this "The evils of A.I.?"
It might be.
I mean, obviously not everything the A.I. manages to draw is 100% based on my art. I'm still prompting with text, and the A.I.'s only using my Derek/Greenie character as a base. The rest is from the world of the Flux model.
Sometimes I'll even see signatures pop up on the corners of images! It's kinda alarming when it does things like that, and is a reminder of just how little the A.I.'s taking from me. .. and how much the A.I.'s taking from elsewhere.

Hmmm..

I've also started using the various AI generators to make the "CoverArt" for my games, because my low-res pixelart is bad enough! Me trying to draw high-quality cover art is a whole different kettle of fish.
... And even my best attempts used to end up looking like kettles of fish!

I mean, which would you rather have...?

or

Yeah, though my sprites are usually things I've worked on, I've kinda opted to just let the A.I. take over when doing hi-res stuff.

.. Lazy Jay!

5. Music - Yep...

I've created almost 1,300 tracks over the years, but as 2024 came to a close, and the A.I. got more and more powerful, it's gotten to the point where I'm starting to feel redundant, as far as making "ALChoons" goes.

Instead, then, I've started putting small A.I. Generated songs in the end of my blogs, each day.
I usually write a bunch of lyrics (or get ChatGPT to help) and then feed Suno with the lyrics, alongside a snippet of one of my many tracks. The two elements combine rather well, and usually end up producing something unexpectedly silly.

In terms of what I use, it's usually Suno.
Udio can be great for doing instrumentals, but I've not gotten along with its vocal stuff.
Sunauto's a recent addition to the list, and seems to come out with its own curious style of music.
But in general, I tend to stick to Suno. All three will give you free daily credits, so you can typically make about 10 tracks a day, depending on which you're using. I've subscribed to the lower tier of Suno, so I can do the uploads and whatnot, but if you're simply playing around and experimenting, the three of them together oughta give you some sort of decent quality output.

Just be aware of copyrights, and each one will usually ask you to always include a link-back at the very least.

Suno's staggeringly good, though, and with the recent v4.0, the overall quality has become so much better than it was just a couple of years ago.
We've gone from slushy mush to high quality audio in an amazingly fast amount of time. I can't wait to see where it goes from here.


A.I. generated music is very much on the "oooh, that could be copyrighted..." fringe of the A.I. world, though. It's definitely getting cues from "somewhere". And sometimes you can kinda hear it, off in the background. Little snippets that you recognise from some long-forgotten song you heard somewhere.
I feel a little less worried about this because I'm also feeding it snippets of my own music, and it's riffing on those instead of whatever's in its training data.
My music, My lyrics, but then it goes off on its own adventure. That's ok, right?! ...Maybe.

And with over a thousand bits of music that I've made, I've got plenty of options to play with.
It's still likely doing copyright theft galore as it goes, but I feel about 3% less guilty, knowing it started with me as a baseline!

As for my own music creativity.. That's pretty much been abandoned now. The A.I.'s gotten TOO good when compared to my own output.

Does this mean I've retired, as far as the music goes?
Yeah, I guess it kinda does.
And that's somewhat alarming.
I haven't really done any music at all, this year. A little ditty here or there. A jingle, an outro for videos, but that's about it.

The A.I.'s overtaken me by leaps and bounds, but as disheartening as it is, I am also enjoying the output that the A.I.'s creating.
...
Is that how the rest of AGameAWeek's going to go?

Bit by bit, the A.I. taking over all my little things that I do, until there's nothing left for me?
We're already at the point where people are getting the various A.I.'s to make versions of Snake and Breakout without breaking a sweat. Give them a few AGameAWeek styled prompts, and I bet I'm out of a fictional job that doesn't even exist, within just a year or so. Maybe even within a few months, at the rate the A.I's are growing.

Hmmm..

6. Sitcom - Yep.. Yeah yeah yep...

Back in 2000 (literally 2000, not like "The 2000's" but actually the year 2000) I decided to write a silly mini sitcom. A short scene or two each day, and then popped it online.
It didn't last long, and I gave up after about 150 or so days, when it became evident that nobody ever read it, and I was just writing it to amuse myself!!

Back in June of 2024 (7 months ago!!?) I fed all the old episodes into ChatGPT and asked it to write me a new episode, and the output wasn't particularly great. It didn't get the humour all that well, and wasn't particularly strong at doing callbacks, instead doing references with links to where it got the references from!!

I did a rewrite, and posted it to the blog for LOLs.
And then I did it the next day... and again and again..
It's been a while, now, and every blog since then has had a small stupid little scene, where Dave and Greenie adventure in their obscure world of chaos. (click Reveal under the daily picture of Grenie)

Every day I'll prompt the A.I. with a vague plot for the day's episode. ChatGPT will write an vague scene, and then I'll rewrite it, and give it back to ChatGPT for any additional thoughts.
Between the two of us, we've been managing a scene or two every day for the past 8 months, and Dave and Greenie's had plenty of adventures in that time.

Am I worried about Copyright Theft there? Not as much as I probably should.
It's been very evident from the start that ChatGPT's been writing based almost entirely from the old episodes. Even to this day, it'll keep bringing up "like that time you were in hospital" references, and "like that time you were on Star Trek".. ..!!!
.. Very old episodes!

The humour's very much my style, and with my daily rewrites guiding the A.I., it still feels like my writing style. It's also not coming up with the plots of the episodes. The random incidents that Dave and Greenie get into. Those are still based on my prompts each morning.
Does that make it "OK"?
I dunno. Maybe not.
But the A.I. Corner segment of the daily blog is segmented for that reason.

Daily music, daily picture, daily sitcom scene. All somewhat mostly A.I. generated.
It's the corner of the content that isn't "really" mine.. Even though it's based on what I've done in the past, it's not really "my" work.
But it's a bit of fun each day.
Right?!

Conclusion?

As for the rest of AGameAWeek. Don't worry. For now it's all my stuff. I'm still drawing sprites, I'm still coding games, and I'm still having to try to come up with a myriad of ideas at a rather alarming rate.

But that doesn't mean the A.I. isn't slowly taking over.
A couple of years ago, there was no A.I. corner on the blogs, and I was desperately trying to draw CoverArt myself, making a right pigs ear of it all.

I make this content mostly because I'm bored, but it's hard not to look at what the A.I. can do in literally seconds, compare it to what takes me a week to do, and know fully well that it's getting to the point where the A.I.'s going to take over.

I know this is the real-world worry for a lot of people, and it's absolutely a real worry. It definitely isn't something that's not rapidly happening. The A.I.'s getting really good at this stuff, and with a little well thought out prompting, the future for musicians, game devs, script writers, artists and even actors and special effect artists.. They're all going to be taken over.

We need to think ahead, and rethink the sorts of roles we'll all be playing in the future.

I know that, should my "world" of creativity come tumbling down, then I can at least enjoy prompting the A.I.'s to do my bidding.
If I can spend a few minutes prompting the sort of game I already make, then I can only imagine what sort of grand games I might be able to prompt into existence, if I spend a week working on the back and forth prompts.

As an avid Trekkie, it brings to mind the world of Holodecks. You can step onto a Holodeck and give a few prompts, before being able to live out a game/story/world within an entire 3D reality.
There's also a world of Holo-Fiction in Trek, where creative people (Tom Paris & Voyager's Doctor) would create entire worlds of fiction that people could play out and enjoy.
I mean, obviously we've no Holodecks yet! ! Obviously..

But we've got VR. Could we soon be at the point where you put a headset on, and ask for a virtual world that you can wander around in?
Let's be honest.. We're not that far off that.
There's things like Blockade AI Skyboxes and other such AI's that can generate the illusion of a world, and there's the foundations of tools that can generate that sort of thing as a 3D Unreal-engine based landscape.
This stuff's actually coming together now, in a way that I certainly couldn't have imagined just a few years ago.

I know that there are naysayers who don't think this is happening, that "it's just an LLM, it can't come up with amazing things itself"
But maybe it's not about it coming up with things itself.
Maybe it's more about using our guidance wisely, to create brand new experiences and generations that are more unique than anything we can come up with just on our own?

And I suppose that's what's going on here at AGameAWeek. For the past couple of decades I've near enough been doing this all on my own. Now there's someone, or some"thing" helping me out, and .. hopefully it means I'll gradually have more and more time doing the thing that I'm good at.
...
The thing that the A.I.'s going to be taking over, soon enough, but that I still enjoy doing.

Making AGameAWeek!

Either way, I hope you're still enjoying the output, whether it's mine or not.
Hopefully it's still mostly my content you're coming here for!
I'll let you know, the day I ask ChatGPT to make an entire game, and lazily post that as if it's AGameAWeek!

A.I. Corner

Lyrics : By me
Sound Imported : Walking Organically
> Reveal 🔎

: Download | Suno Link

Alternative version
: Download | Suno Link
Sung by Suno


"Cartoon Derek feeds the A.I. Robot with a bag of silicon chips." by Replicate/Flux

> Reveal 🔎