Doomworld - AKA Troonworld, a dedicated safe space for pedophiles and megalomaniacs - Corrupting the Doom community since 2016

[StealingElectricity]

I recommend you play them with Woof.

Might be a retarded question, but why Woof and not GZDoom or at least Nugget?

 

[Zacharin X]

I like Doomkid. He makes really cool retro wads that are Boom compatible. I recommend you play them with Woof.

Doomkid makes vanilla-compatible wads, that can be played without the use of source ports.

Might be a retarded question, but why Woof and not GZDoom or at least Nugget?

You don't have to play it with Woof or Nugget, GZDoom is fine if that's your preference. People recommend ports like DSDA, Woof, Crispy, etc, because they are closer to the vanilla purist experience, though not as much as something like Chocolate Doom. It's bullshit really, if you're not a speedrunner, you have no reason to switch ports. Another reason might be potential performance issues. If you play some overdetailed, giant abomination, GZDoom might struggle with keeping a smooth frame rate, while with the other ports you won't have such problems.

 

[Shallow Wave]

He actually works in a leisure/sport center as a receptionist

Dudes entire job is to give passes and punch tickets and he probably earns more than me. I doubt there are much Karens in planet fitness to complain about the colour of the weights or something like that. Whats so stressful, issues with his body image? He probably seen lard asses get fit in 3 months behind the desk so theres enough motivation to get him going, so whats the problem?

my military-time-self. I mean just as self-loathing and egoistic, lost, angered as I was.

Wasnt there a dude calling you out on suicide baiting recently?

 

[Perfect Paz]

There is this 'Doom Awards' that started last year to 'counter' the Cacowards by letting people vote for the wads instead of KKKlub but it's not really any different. Wads from the KKKlub still get the most votes plus, allegedly 'Doom Awards' are using AI to write huge texts, so..... no, we really don't have any good alternatives.

That's unfortunate. I'm explicitly a fan of generative AI, but I've been moving away from LLMs lately because I don't like their output. Maybe they'll get better in time.

That being said, structuring it as a matter of popularity is natural (it's possible to get results you don't like and you're bound by them), as opposed to arbitrary (you just decide which one is "best" based on how triggered you are) so I approve of this. I wouldn't fundamentally object to the use of generative AI, since I don't have an ideological objection to it (I use it to great effect to generate illustrations in many styles, with only a few noteworthy flaws to fix remaining), but I can't imagine how one would do so effectively in the case of this since you need to review maps, which requires you to play the game.

I'd probably try to pick this one up myself if I didn't have my own passion projects I'm working on.

EDIT: now that I think on it, I do publish a magazine out of my garage named IDDQD Magazine. It's a variety magazine with a focus on /tg/ stuff. Not to shill too hard, but it's https://iddqd.pub/. We do use AI-generated images and occasionally text, and write articles on things like prompting and fine-tuning, but we also have conventional artists who illustrate or paint for us as well. It's a group effort. So we're openly pro-AI but also openly skeptical of it due to its limitations.

And we may be willing to put new releases of Doomshit into print, where we just print some screenshots and the description of the WAD. But the thing is, I wouldn't want to rely on an LLM to generate a description of said WAD, because I literally CANNOT get that LLM to understand the context of the map unless I feed that information into it.

But then there's also the issue of what's the point of putting to print what is already available online for free, and print might not be the best medium for it?

I'm legit going to be sleeping on this for a bit.

 

[IKOL]

Wasnt there a dude calling you out on suicide baiting recently?

No.

Lemme ask, since when the phrase "I'll achieve happiness or die trying" become a suicide bait? What niggercattle logic some of you autistic cunts use to put such a message behind the phrase?

 

[Shallow Wave]

We do use AI-generated images

Yea it shows, wouldnt hurt to touch them up a bit or use better embeddings because hands are fucky and eyes are weird. AI can do hands pretty well now!

No.

Lemme ask, since when the phrase "I'll achieve happiness or die trying" become a suicide bait? What niggercattle logic some of you autistic cunts use to put such a message behind the phrase?

Cant really judge by the out of context phrases but you are angry all the time so i asked

 

[Perfect Paz]

Yea it shows, wouldnt hurt to touch them up a bit or use better embeddings because hands are fucky and eyes are weird. AI can do hands pretty well now!

We're getting better about that but I'm also going to get a cash injection in the next year or two from taking on another second job so we'll likely have more commissions. The biggest criticism of our AI-generated images isn't hands or eyes. It's signage having shitty language, not rendering right at all.

Touching them up by hand might be worthwhile as well. My fiance is an illustrator but I have kept her hands off of this project because it's overall very male-coded and I don't want to burden her with my hobbies. I don't want to twist her arm for my aspie shit. Although she has offered to stick her hand in it, it comes off as being about wanting to be good to me and not about wanting to realize my "vision" just because it's a good one, so it feels wrong. I feel less wrong about commissions because I'm paying them and less bad about AI-generated artwork because it's just GPU cycles.

EDIT: I should mention that this only sells a few dozen copies each month, but it's more than what I spend on it, so I will be filing it as in the black to the IRS, which I find to be an accomplishment in this economy.

 

[StealingElectricity]

And we may be willing to put new releases of Doomshit into print, where we just print some screenshots and the description of the WAD. But the thing is, I wouldn't want to rely on an LLM to generate a description of said WAD, because I literally CANNOT get that LLM to understand the context of the map unless I feed that information into it.

Very interesting, though the mind shutter: how do you even train this thing? What data do you feed it? Besides the merits from such AI seem to be minuscule, though the opposite seems interesting. Wonder if we can get it to do the reverse - generating maps from said descriptions. We have a rather large WAD database to feed it, so it shouldn't really be a problem in the near future.

 

[Perfect Paz]

Very interesting, though the mind shutter: how do you even train this thing? What data do you feed it? Besides the merits from such AI seem to be minuscule, though the opposite seems interesting. Wonder if we can get it to do the reverse - generating maps from said descriptions. We have a rather large WAD database to feed it, so it shouldn't really be a problem in the near future.

I've fine-tuned Llama2 on TTRPG manuals. The merits of such AI are in things that are relatively formulaic. For example, when writing a new setting in the Narration System we have, we need to come up with about 20-30 Proficiencies (such as Brawling, Lore, or Haggling; think Skills). There's a certain format they follow, where each one has an illustration, a scenario of the Proficiency in real use, a couple of paragraphs describing what it does and what it's about, and a description of different roll results and how to interpret them. We used an LLM to generate this text.

However, we didn't do it all in one shot. We used the LLM to expound. So we told it what the scenario is, etc in the prompt: ("write one for Brawling where the scenario is a big lug in an alleyway slams two muggers' heads together, here's the short sentence for describing the Proficiency") The format was already in the context window. Then we'd proofread it, edit it, and put it in.

However, after all that, if you turn the page, it's a description of what we went through to get it to be worth a shit. See, the LLMs we used loved to use a certain sentence structure over and over again. "In the cold, dark streets of Degeneration NOAH." "In the unforgiving shadow of Degeneration NOAH." And so on. It was so egregious that it did this every fucking time, even if we explicitly prompted it not to. That use of language, and a few others, ended up getting edited out of it.

But even after that editing process, even with the time spent fine-tuning, it was still a more efficient use of our time to have these be written partly by LLMs. And the same is true for images. A lot of these images, the ones that have better eyes/hands and better signage text, were heavily edited. So on the one hand, generative AI helps us get content out faster, but on the other, we acknowledge its limitations and don't overrely on it, or try not to. To us, it's a productivity tool, not a replacement for the human, and since we got in on that line of thinking early, you're going to see a lot of really goofy flaws, especially in earlier issues. We were there in January 2023. We rushed out the gate bragging about our use of generative AI because we regard it as forward-thinking.

On the subject of LLMs for procedural map generation: probably not a good idea. See, LLMs under the hood are text completion models with a dash of randomness thrown in. They are given text, and then are designed to complete it. So given "the quick brown fox" it might return "jumps over the lazy dog" or something similar. But thanks to the randomness element -- which is literally called "temperature", where the hotter it is the more random it becomes -- it might produce something entirely different. The ones we have are designed to respond to things phrased using the imperative tone, that is, instructions and commands. It produces text to meet the instructions. And the training data is literally language.

So what we would have do to get it to procedural map generation is translate ALL OF THE MAPS themselves into a human-readable language, feed them ALL into a new LLM trained on literally nothing but them, and then just let it run based on a few small sectors or something as a prompt. Then it MIGHT produce something somewhat coherent, but might also produce maps that literally can't be beaten because the end switch is in an impossible-to-access sector or something like that. So it'll still need human editing.

If you want an old project that does procedural map generation that isn't trash, look at SLIGE.

Here's a better use of generative AI for Doomshit: generating assets like textures and monsters. Fine-tune Stable Diffusion on 1990s-era FPS games in general, not just Doom, and then you can have it generate pixeltastic wall textures or something.

Then from there you can use a technique that rap/electronic producers call "digging." You just hit the random button, essentially. A simple prompt that gets you something vaguely like what you want, and you turn the temperature up to like 1.6 where it normally might just be 1.0. You hit that random button a lot. Over and over again. Then you eventually get something really good, at least conceptually, even if it's fucked up. You polish that turd. That's your production asset.

You will spend about 50% of the time to make it entirely from scratch. You'll spend real time. But not as much. Then you let the haters hate because you made something cool and they didn't.

 

[IKOL]

Cant really judge by the out of context phrases but you are angry all the time so i asked

My brother in Kiwi Christ, are you aware that I'm Russian? If not, you are now.

 

[StealingElectricity]

See, LLMs under the hood are text completion models with a dash of randomness thrown in.

I mean, isn't that how the brain processes information too? I have made some basic image recognition AI, and as far as I can see, the neural network pretty much does the same thing as the brain in terms of generating output. Most of the stuff we create are based on already established narratives that we subconsciously consume, but I digress.

Still, given that our current best map generation system is Oblige(which I think SLIGE was based on)/OBSIDIAN (which, as far as my understanding goes, splice random map rooms from presets), which I haven't tested yet, tough I generally hear that they are meh at best.

Here's a better use of generative AI for Doomshit: generating assets like textures and monsters. Fine-tune Stable Diffusion on 1990s-era FPS games in general, not just Doom, and then you can have it generate pixeltastic wall textures or something.

Actually a great idea, not sure why I haven't thought of that yet - very clever.

 

[Perfect Paz]

I mean, isn't that how the brain processes information too? I have made some basic image recognition AI, and as far as I can see, the neural network pretty much does the same thing as the brain in terms of generating output. Most of the stuff we create are based on already established narratives that we subconsciously consume, but I digress.

Still, given that our current best map generation system is Oblige(which I think SLIGE was based on)/OBSIDIAN (which, as far as my understanding goes, splice random map rooms from presets), which I haven't tested yet, tough I generally hear that they are meh at best.


Actually a great idea, not sure why I haven't thought of that yet - very clever.

Neural networks do work similarly to the brain, yes, that's where they got their name from. It's a bit simplified but yeah, you're on the right track. You could maybe also do monster, weapon, and projectile assets. I'm not sure if fine-tuning all of them into a single job would be better than doing them as separate fine-tunes but that's for someone more interested in that experiment than me.

Anyway, I know I'm harping on about this, but, generative AI really is the future, and 2023-era goofy assets aren't its downfall. Its downfall, if it has one, will be people giving oxygen to the idea that generative AI is inherently bad by pretending they aren't using it. The duplicitous behavior implies that the anti-AI people are ultimately correct that AI-generated content automatically lacks sovl or whatever, which they aren't.

 

[Shallow Wave]

how do you even train this thing?

If you arent a nerd you take stable diffusion GUI from git, get bunch of pictures you want to train AI on, write a text file with danbooru style tags for said pictures in it, then press the "train" button in SDGUI

 

[HumptiDouji]

What does everyone here think of his megawad Jenesis? I replayed it and while it looks decent to the eyes IMO, it likes to spam damaging floors like there's no tomorrow.

I played it when it was new, and I seem to recall thinking it was overall quite decent. Not perfect, but enjoyable and that it didn't feel like a waste of time, but my memory of it is kinda fuzzy by now. It relied too much on hurtfloors, that part is pretty familiar to me still. I enjoyed it more than I enjoyed MyHouse.wad

For his music, it doesn't blow my mind at every turn or anything, but I can't think of any work of his which I've outright disliked, some of it is really good, but a lot of it is also forgettable. By forgettable I mean tracks which are serviceable for whichever level they are on, but which don't stick in your mind after that.

He's bri*ish doomkid

Has he had the same kind of deranged freakouts as Doomkid?
The picture that gets painted here is a guy who's talented, yet overrated in some ways, and who's lonely and depressed IRL, so by the standards of this thread he could be doing a fuck of a lot worse.

Also, found this gem. It's kind of old but still relevant enough to post it.

If he's reading this, he should know that online dating services are actually a really horrible way to meet people.
Try your luck with ANY other approach.

 

[IKOL]

Has he had the same kind of deranged freakouts as Doomkid?

Does attempt to come out with dragonfly counts? Oh wait we live in 2024, nevermind, carry on...

 

[Shallow Wave]

Does attempt to come out with dragonfly counts? Oh wait we live in 2024, nevermind, carry on...

Dragonfly was engaged then, no way

 

[StealingElectricity]

If you arent a nerd you take stable diffusion GUI from git, get bunch of pictures you want to train AI on, write a text file with danbooru style tags for said pictures in it, then press the "train" button in SDGUI

Sure, but how can we apply this to Doom WADs? How can we instruct the AI to parse the WAD and derive context from its data? Or maybe the WAD must first be parsed and then have specific sections of it fed to said AI? Either way, sounds painful.

 

[Shallow Wave]

Sure, but how can we apply this to Doom WADs? How can we instruct the AI to parse the WAD and derive context from its data? Or maybe the WAD must first be parsed and then have specific sections of it fed to said AI? Either way, sounds painful.

Well in ideal world you could just feed whatever learning algorithms you using contents of LINEDEFS, SIDEDEFS, THINGS and SECTORS of a limit removing map and let it run for a while searching for patterns. Its hard, you need to write custom code and data is noisy. In less than ideal world you can grab udb, screenshot map layouts and feed it to stable diffusion so it can gen you a map or a room layout. This is easy, funny and guaranteed to make map makers seethe

 

[Zacharin X]

Since there's nothing else to talk about, let's change the subject a little bit. To all mappers in this thread, let me ask you, have you ever tried to seek playtesting in Austinado's abortion of a thread? If you did, please, share your experiences with it. And to all the would be mappers in this thread, let this be a warning to you, so you don't waste your time.

https://www.doomworld.com/forum/topic/118850-who-can-test-your-maps-list-of-active-playtesters/

List of active playtesters, my ass! I have nothing agains Austinado, i've came across a few of his Doom streams in the past year, and while i usually don't stick around to watch him, from what i've seen, he seems like a stand-up guy. I also get the idea behind this thread, on paper it's great, but it's poorly executed and badly maintained. Here's the problem - probably half of the listed "playtesters" are dead weight, because they're either logging in every once in a while, or they no longer visit Troonworld. Then you have the other, who are suposedly active forum users, so what do you do? Do you PM one of them? Or two? Or all of them? Either way, you're in for a disappointment. This is what's going to happen... most of the testers who you'll PM, won't reply to you at all, even if they're clearly logging in on the forum and posting in various threads on a daily basis, they'll just ghost you. Most of these cunts, are some good for nothing noobs, who want to be seen as useful, but they end up biting more than they can chew, and once the requests start coming in, they quickly regret signing in to the idea of testing. One retard once told me, "Well, you have to consider that playtesters are still people, dude". To which i replied, "Hey, dumbass, you sign up for this, i didn't force you to put your name in a playtesters thread, you did it yourself. If you don't want to playtest, then remove your fucking name from the list and stop wasting everyone's time". I forgot who he was, would've been interesting to see if he's still listed as an active tester.

Other testers will reply and tell you they'll check your map, but they'll never return with any feedback. Some will actually play your map, but their feedback will be underwhelming... like one sentece that tells you nothing of their experience with it. And if you're lucky you might come across one or two testers, who'll give you good advice and interesting suggestions... or maybe even upload a video with commentary, or a demo for you to watch. Out of those, probably the RAMP furry DavidN is your best bet, but there's a catch with him - because he's the best of the bunch, he gets so many requests, that you'll have to wait for months. And by the time he gets to your map, it'll be too late... you'll either have it publicly released, or completely revamped, so his playtest will be somewhat obsolete. Another good one is Kan3. Like DavidN, Kan3 aslo records videos with commentary, but he's much faster, maybe because he doesn't have many requests. The only downside with both of them is that they play on GZDoom with freelook on, and Kan3 in particular uses the freelook to break and cheese a lot of encounters designed around the autoaim... it can be infuriating to watch at times, but he's a nice guy and that compesates for his bad gameplay. Biodegradable also records videos, but also ghosts requests on PM, and i've seen people calling him on it. He's also a complete retard, so don't bother with that fag.

Overall, i have to say, Austinado has dropped the ball with this playtesting thread, but he's not the only one who can maintain it. Any troon moderator can do that job, but of course the fags wont lift a finger. I think the whole process is fucked, which is unfortunate, because there is a simple solution. The process will be a lot better if the thread was open for direct requests, so that anyone can post their maps there, and then the playtesters who are willing to play them, will give feedback in the thread, and everyone can see who is actively participating and who doesn't. If the mapper wants to keep his map private, he can still PM the testers who are giving the best and most helpful feedback. This will also undoubtedly reduce the solo "my first map" shitposting in the wads subforum. Most importantly it won't force mappers to be PM-ing 20-30 different people, half of which will ghost them and make them feel like fucking bitches for asking nicely.

 

[Shallow Wave]

just ask your fried to playtest your map ez