Toggle sidebar
Front-page

AI-assisted recomps and decomps - where do you draw the line?

AI-assisted recomps and decomps - where do you draw the line?

You'd have to be living under a rock to have not noticed the recent boom in static recompilation and decompilation projects - it seems like every day, multiple projects are announced; from popular titles to more obscure games with niche audiences; at the rate they're appearing it feels like it's only a matter of time until every game is playable on every platform.

For example, just recently The Legend of Zelda: The Wind Waker has made its way to Windows and Mac natively - bringing with it high frame rates, widescreen support, improved rendering, HD texture pack support and more... AstroCore has brought Super Mario Galaxy to PC, gen1recomp has added FireRed and LeafGreen, the list goes on.

It's no surprise to anyone that this boom in projects has been almost fully in part to the rise of AI assisted development or in some cases, projects that are entirely vibe-coded.

Many of these projects disclose outright the fact they were built with help from large language models such as Claude Code. For example, The Wind Waker recomps readme says it was developed "with substantial AI assistance for code, testing, documentation and debugging".

So here's the question, beyond the ethical and environmental impacts; these recomps and decomps cost nothing. They’re made in part by people in their spare time and can provide easy access to games that are perhaps stuck on dead or hard to find hardware, run or perform badly on their original platforms; or that the original publisher may have delisted entirely. Yet as soon as an LLM shows up in the credits a big part of the reaction devolves to "AI = bad", whatever the resulting product may look like. Of course, some of that criticism is absolutely well founded: people worry about undocumented, unoptimized and spiraling codebases; about the craft and passion of hand reverse-engineering and about how these models were trained in the first place.

So who's really right? Is it the people who see AI-assisted ports as a shortcut that cheapens the scene, or the people who think a free, working, improved and accessible version of a game they love is worth having however it was made?

Where do you personally draw the line?

Back in August, Scarlet's community check-in on how we cover AI-assisted projects garnered more than 360 replies. Scarlet suggested that when a project we feature uses AI and makes it clear that it does; we will endeavor to inform you depending on of course on the information publically available and to the extent that you should expect from an independent website providing simple news coverage. Giving anyone who would rather avoid these projects a clear indicator, without anyone having to dig through repositories for hints themselves.

So let us know:
  • Where do you draw the line on using LLMs in such recomp and decomp projects?
  • Would you play a game knowing much of its port was written with AI? Or would you prefer to avoid it?
  • If an AI-assisted recomp were the only way left to play a game you cherish would you take it? Or would your principles keep you waiting for an official remaster?
As always, keep it civil please...
 
  • Like
  • Love
  • Haha
Reactions: console, Night007, bonkmaykr and 18 others
Click to expand...
AxlSt00pid said:
My brother in fucking christ, when did I ever mention Wind Waker HD for Switch 1

My experience is playing 3 recomps where 99% of things were done by AI and they were a stuttering mess: A Tekken 3 one where the latest release halved the framerate, a MK Super Circuit one where it had quite some frametime and visual issues, and a Tony Hawk Project 8 recomp where the shader compilation stutters made the game completely unplayable
Click to expand...

So you based your entire opinion on 3 recomps? You're aware that most handwritten emulators takes many years to get to a somewhat usable state but that doesn't stop people from signing their praises.
Just look at PS4 and PS5 emulation. No one is calling that slop even though they are in a worse state for the most part and have been in development for much longer?

The problem here is that you test something that has not been out for very long and expect it to run flawlessly.
Post automatically merged:

isoboy said:
That's their problem, not mine.
Click to expand...
Indeed
 
As sometime with the actual skills to do this kind of work, I like it. For this specific purpose anyhow. There is no job market for recomp and decomp work. We're also seeing a surge in them now because AI helps get people through the hardest part.

As long as they are taking time to manually and polish translations, images, textures packs, etc., I think AI is actually harmlesdly enabling tons of creative people in the gaming community that lacks the technical skills.
 
  • Like
Reactions: The Real Jdbye and Octoblimp
I don't think there is a line personally. Why? Because idgaf if a project uses AI or how. I would prefer its use be disclosed but I'm not like a nazi about it who's gonna flip my mind if I find out a project used AI and didn't disclose it. Given how a lot of people in the homebrew scene (and I'm talking everywhere, I'm not picking on anybody specifically; This includes subreddits etc.) have been acting towards projects that do use AI, even for the tiniest thing just immediately calling it slop and vibe coded and refusing to use it even if it is actually quality and decent, I wouldn't blame them if they didn't want to disclose it.

Me? I will always disclose AI use as its a reasonable and responsible thing to do. I also think this quote resonates with me quite a bit.

NewLoveMachine said:
It's the wild west right now, some people are taking it slow and easy while others are churning shit out at a rapid pace. The pace seems almost performative, like "look how fast I can get this vibe-coded". Cool, but does it run well? Provide any benefits original hardware or emulation does not?

At the end of the day I'm deciding if I want to waste my time with the final product, and really all I care about is honesty and quality. It speaks to your character if you're taking full credit for something you might not have put much effort into and raises the question on what else you might be lying about. Slow down, it's not a race.
Click to expand...
People are trying to become the next PabloMK7 or some shi like that, and whilst even if you use AI its possible, its definitely not happening if you rush it. I've definitely been seeing a lot of rushing, and those that are rushing to get something public (much like the MKWii project did) kinda I think give us AI coders (yes I'm calling vibe coders AI coders, I prefer it) a bad rap. I used Claude to port Android 2.0 Eclair to the 3DS, and even though it was publicly launched as version Alpha 1, it is most certainly not that early of a state. Userspace works, WiFi works, Wireless ADB works and so much more. It took me 3-4 months to feel comfortable enough to release a public build of it and announce it to the world.

That's what people should be doing. Taking their time, iterating, testing etc. until you'd be like "If I saw this today made by somebody else in this state, I'd wanna try it out." Notably, the same day my Android build launched to the public, was the same day just a couple hours prior that another Android project had been announced. Note the keyword "announced." It wasn't released, no source code, no compiled files. Just a couple of photos that honestly I had a hard time believing. The OP claimed they had only worked on it for a week, which, speaking as somebody who did actually port Android to the 3DS, aint no chance in hell you ported Android to the 3DS in a week, let alone get it to run games that are playable. And this is part of what I mean by certain individuals giving AI coders a bad name. They even released their source code then admitted it "doesn't compile" like bro are we fr rn?

What I'm trying to say is if people just took their time, released quality stuff then AI coders wouldn't be so bad. But due to how accessible AI is, we will likely continue to see these low quality slops put out by people who really just don't care. My philosophy is, don't release a project til its ready. I'm working on another project right now after having launched Android and that wont launch, nor let alone be announced until it is 100% completed.
 
  • Like
Reactions: The Real Jdbye
I think it’s fine as long as there’s full transparency that it’s built with ai. I’m interested to see this large uptick in recomps coming out simply to shake the tree and see where the legal limits are that Nintendo and co would theoretically be willing to go to at least attempt to shut this down, even if they technically shouldn’t be able to do anything about these recomps. Hopefully just get some precedence in the books
 
MariArch said:
I think it’s fine as long as there’s full transparency that it’s built with ai. I’m interested to see this large uptick in recomps coming out simply to shake the tree and see where the legal limits are that Nintendo and co would theoretically be willing to go to at least attempt to shut this down, even if they technically shouldn’t be able to do anything about these recomps. Hopefully just get some precedence in the books
Click to expand...

I mean as far as I know it's not illegal to paint a copy of Mona Lisa and give it to a friend after telling it that you were the one painting it?

As long as you are not using the original work or distributing copyrighted assets then I don't really know what they can do.

Think about what OpenMW is. It's a community developed game-engine for running Morrowind.
You have to provide the game data yourself. And I feel that this is pretty much the same as the Recomps and Decomps. (at least the ones that does it the right way)

I don't think there is much that can be done to be honest, all the tools are available to train and fine-tune your own LLM's and run them locally already so even if all the ChatGPT's and Claude's are being enforced to block reverse engineering of this kind then it will still live on. Fighting it seems like a hopeless fight to be honest.
 
  • Like
Reactions: bonkmaykr and Octoblimp
I'm all for it. It doesn't matter at all. What's great about this is the coding methods can help coders discover new ways to do things quicker and easier as well as methods never even thought of. It's a win win for us all.
 
  • Love
Reactions: Octoblimp
Hey, I never expected my Switch port of DK64: Recompiled to blow up like this, so here are my two cents on the matter.

First, to be clear: the port was written with Claude Opus 5.5. My part was deciding what to build, guiding the work, testing every build on real hardware, reporting the bugs and deciding what ships. The README says it openly: https://github.com/Trinalord777/DK64Recomp-NX#how-it-was-made-ai-use

Honestly though, I just wanted to play DK64 natively on my Switch, and I was happy to share it on GitHub and here so others could too. I know a lot of new decomps and recomps are now made with heavy use of AI, and I understand what that means for developers who spent years mastering low-level C and assembly. Still, I think AI used reasonably, to make old games more accessible, is better than not being able to play them at all.

In theory, I think AI can be a valuable technology; it all depends on the practice. Projects that would take months or years can now be built in days or weeks. It can also help with debugging that would be much harder to track down by hand, and I learned far more doing this project than I would have otherwise. That doesn’t mean the result can’t be messy, buggy or unplayable: it still needs a human to test and verify everything. But it gives you a working base that would otherwise take endless trial and error. That said, I have a lot of respect for the people who decomp and recomp from scratch, and I understand the worries about this new technology.

More broadly, I’m against AI pushed everywhere, at any speed, with no rules. Today it mostly concentrates power and money in a few big companies, while everyone else pays for the new data centers through their bills. To me the main problem is unregulated AI, and companies racing to collect all your data, sell your information and make you a dependent customer. That didn’t start with AI; it goes back to cookies in the 90s. What’s worse today is politicians who actively push for more of this, more consumption and higher bills for people who never had a say.

If you’d rather avoid AI-assisted projects, I completely get it, and that’s exactly why I say it up front. But I think It can bring a lot of good things, as long it used in the right way
 
  • Like
  • Love
Reactions: RedoLane, eyeliner, The Real Jdbye and 2 others
i have ported the three gba fire emblem to pc ,added full documentation of every fiunction ,sprite and options inside them.added tactile options as a third choice of control.Native PC port.And all this out of spite.So yeah AI helps alot.despite the haters
 
  • Like
Reactions: eyeliner and Octoblimp
I don't really see the point of recomps/decomps games. To me, running games on emulators are already very well optimized and playing, for example, zelda Wind Waker HD on CEMU with graphics mods or on Windows directly is the same
 
Keylogger said:
I don't really see the point of recomps/decomps games. To me, running games on emulators are already very well optimized and playing, for example, zelda Wind Waker HD on CEMU with graphics mods or on Windows directly is the same
Click to expand...
You don't see the point of decomps?!? I can understand recomps because...well, to me at least recomps aren't as useful as decomps. Like, recomps are nice as a substitution to have to avoid decomps but decomps provide so much greater flexibility in things like modding and makes porting to other systems easier. That's what decomps are great for. Porting games to systems that otherwise can't emulate them or emulates them rather poorly. For example, the 3DS can't emulate the PS2, so decomping and porting a PS2 game natively to the 3DS can make it playable because its not emulated.
 
  • Like
Reactions: AlexMCS
Keylogger said:
I don't really see the point of recomps/decomps games. To me, running games on emulators are already very well optimized and playing, for example, zelda Wind Waker HD on CEMU with graphics mods or on Windows directly is the same
Click to expand...
N64 games on emulators were a big mess for a long time related to graph-audio hardware properties. Recomp/decomps entirely removed the problem while adding technical enhancements that were previously hard-to-attain.
 
Logged in for first time in forever to chip in my opinion:

AI is a Tool. If it is ASSISTING the Human and not completely replacing their efforts, that's where the line stands for me. The monotony of doing the same thing 1,000 times? Cut it out. Do the work once and have AI manage the fill-in.

If it is doing all the work? That is now officially just entering dubious territory.

Related but not same: I stand on opinion that AI Translation patches are fine, long as there's someone proofreading it all, doing double checks for specific terms/parts to ensure it translated well, and boom.
 
  • Like
Reactions: bonkmaykr, AlexMCS, DarknessPlay3r and 1 other person
Keylogger said:
I don't really see the point of recomps/decomps games. To me, running games on emulators are already very well optimized and playing, for example, zelda Wind Waker HD on CEMU with graphics mods or on Windows directly is the same
Click to expand...
That depends entirely on the platform on which you are wanting to play the game. PC? Yeah, the emulation scene on PC is extremely mature and well developed. Android? A little less so. Switch? Somewhere in between.

The point is to A.) bring a title to a platform that it otherwise might not ever see and B.) include improvements that may not be available otherwise, such as widescreen hacks, multiplayer optimizations and hacks, 60fps mods, graphical improvements (not just scalers, overlays and upscaling but overhauls, native rendering to improve performance, post processing, hud mods, etc.) and in many cases, bringing features that are unavailabile via any other method. This includes touchscreen and gyro support, flipgrip support with new touch enabled menus (giving a ds/3ds style gameplay), randomizers, new game modes, cross save support, localisations, Retro Achievements support, improved controls, improved camera control, and even restored/migrated features from other versions of the game (i.e. a game that was released on SNES, 3DS, PS1, etc. might have different cut scenes and material that appeared in some versions but not others; it is sometimes possible to get all of that material into one version of the game and create an "ultimate" or "reference" version).
 
Personally as long as the port works well... Then who gives a shit? Oh, no! A person didn't spend a fraction of their lives manually naming all these functions and variables. A machine did it in a week instead! The horror! People are so weird about this kind of crap man. And I can understand it when it comes to stuff like OS development where security and maintainability is crucial. But these are unofficial ports of video games. Chill out. Lol.
 
Last edited by Night007,
  • Like
  • Love
Reactions: Octoblimp, AlexMCS and shaunj66
Alright, so I already gave my stance on the matter in a few News threads that are about AI-assisted decomps/recomps/PC ports, but knowing there are actually people here who are a part of this whole wild west, one more time shouldn't hurt.

So yeah, drawing the line. umm...
At first I was very hesitant about vibe-coded recomps because I was much more anti-AI, but after having more experience with Gemini and Claude, I believe it all comes down to this:
Using AI the right way.
Basically, vibe-coding is usually a sign of using AI the wrong way, or at least that's the impression it gives. It's kinda like trying to draw a portrait but not having any skills for the entirety of it, so then it's all about asking an actual artist to draw 90% of the portrait, while the man in charge only draws a funny face. I'm not saying that AI are actual artists lmao, but the method is exactly that: thinking that something else is actually a "someone else" who will do most of the work.
These tools are fine if the human behind it is the hard-worker, but that's just something that needs to be elaborated by, well, the human.

I think communication is very important, and if an AI-assisted project's dev doesn't elaborate on the use of AI, that's when I'm starting to doubt it's most important aspect: Stability.
I recently played a Conker's Bad Fur Day recomp. For the most part it worked greatly, but the aspect ratio(16:10 on steam deck) wasn't 100% aligned no matter what I tinkered with, it crashed a few times in very important moments, and overall it had a few rare-but-weird texture pop-outs. There are a few N64 recomps which use the same N64recomp tool that are more stable than this, which is why I'd call the CBFD recomp: semi-stable.
But then you have all those PS1 recomps that not only do nothing with the aspect ratio AND they come with their unfair share of bugs, to the point they become uncomfortable to run. They're released so quickly and the decomps aren't even finished. I mean, aspect ratio is one thing and it's my bias, but if I want to play it in it's original intended way, I'll just use an emulator.
I guess that's the second most important aspect: Enhance it.
I mean if you have the tools to port it over, how about giving some QoL improvements while at it? Since it's a free-time project.
Better analog controls, hotkeys, having a Graphics settings menu in-game, the kind of things that actually make it more like a PC port.

TL;DR: AI is not automatically bad. If someone wants to recompile an already existing decompilation, then the real exam is all about stability and the extra sauce. It needs to be playable from beginning to end WITHOUT crashing, and if it's a game that originally deserves better, here's a chance to redeem it. Work harder, don't abandon it in a month, and ACCEPT, FEEDBACK, PLEASE.
We're already too late and deep into the AI craze, so I'm going to keep my eye on the smart developers and move my eyes away from the naïve.
 
Last edited by RedoLane,
  • Like
Reactions: bonkmaykr and Octoblimp
I expect that the people who worked on any project have prior engineering experience. If they don't, I don't want to use their software. I don't know what is in it and the experience is likely worse than simply emulating the game or patching it myself.

The problem with trying to cater to anti-AI members of the forum, is that GBATemp editors are probably not programmers and probably don't have time to psycho-analyze the codebase of every release that gets announced. So I propose that you instead look at the history of the lead developer. What do their contributions to other repositories look like, where have they been?

For example, a non-programmer can go to my profile, and my org's profile, and see that:
- Many of my repositories are years old and at least one of them has a development history spanning over a year.
- I have a lot of mirrors and forks which I've made changes to, and a dotfiles repo, so I probably use Git for tracking small patches and not just software development.
- I have smaller programs that aren't announced anywhere or are part of a larger project.
- Many of my commits are small but not overly compact, and are only commented when necessary, parts of it are self explanatory. The commit descriptions appear to be written by a person, sometimes lazily, this is typical of a hobbyist.
- My website occasionally describes game development concepts in a very technical but human way, including descriptions of unreleased projects
- I have projects shared with people that appear to have a connection to me

Whereas a vibe coder will usually:
- Have no history or external connections at all
- Only work on larger projects and not smaller pieces
- Advertise everything
- If it's a decomp, produce code that is remarkably unreadable even to a non-programmer because the AI can't understand the code without context from human documentation (like the Windows API).

This is a lot less detective work than learning to spot code smells.

Chances are high that if I ever used AI, it would be done so scarcely and responsibly, in such a way you wouldn't tell the difference. Software engineers understand the limitations of an LLM and why relying on it is not appropriate.

People with no experience, on the other hand, frequently reach for conveniences without seeing the downsides until they've given up.
 
Last edited by bonkmaykr,
  • Love
Reactions: DragonMals
People really like to belabor how much more "accessible" it makes old games, and trot out the "it's better than nothing" line.

...except we don't have nothing. We have over two, now nearly three decades of emulation research & development behind us. UltraHLE was better than nothing. NESticle was better than nothing. bleem!/bleemcast! (written in Assembly, no less!) were better than nothing. Heck, even your average plugin-tangled N64 emulator is better than nothing.

Most of these recomps are more like lousy middles with the same or worse results than an emulator. Probably because some are preexisting emulators wrapped in an executable with no game-specific fixes or tweaks beyond said emulator's dynamic recompiler already does. They don't get more because some behind these (fool's) gold rush recomps don't care past the initial wave of press.

I'd be willing to accept that people are "using it wrong" if it wasn't for the attitudes of defenders, such as acting like code review from a human isn't the barest of bare minimums or people salivating to make carbon copies of pre-existing work just because the programmer dared having a strong opinion about their own ethics or bar use of incompatible software.

And it strikes me a bit odd how now people are harping on how "it's all for free!!" as if that's stopped people from vilifying free homebrew devs for anything before this. Like, for taking "too long" on a hobby project that doesn't pay them (and they CANNOT monetize due to the lack of legal precedent, unlike work for a commercial emulator system like Polymega or something). Or acting like they didn't ask for help in that time when the entire point of open-sourcing was to crowdsource help... that never came.
Or for holding onto an entrypoint until a wider range of hardware is open to it. The latter of which has DEFINITELY not happened ever since & isn't now a huge tension point for people looking to buy a highly-anticipated game while still having access to unsigned software.
 

Site & Scene News

New Hot Discussed User Submitted

Go to forum More news

Popular threads in this forum