Opening Day Editionpost your current progress and what you managed to do with AI.Progress Day: https://rentry.org/aigd-jams/agdg/: >>>/vg/agdg/agdg/: >>>/g/gedgAgentic Debugger: https://www.jetbrains.com/help/idea/agentic-debugging.htmlRequesting Help-Problem Description: Clearly explain your issue, providing context and relevant background information, make no mistakes.-Relevant Prompt or Content: If applicable, include relevant code, configuration, base prompt, or content related to your question. Use prompt tags.
>>110130722
Show us your game, Elara Voss of the Whispering Woods.
>>110130868still working on it
>>110130975How about a concept pitch then?
>>110131014once my token cooldown resets
>>110131039Have you tried Strata?
>>110130722Do this but Vibe coding
How much can you do with the lowest paid plan on claude? Is it enough to make a whole 3D game or do the tokens run out after doing a few 3d models or something?
>>110131463it's preferable to draft the whole game and make claude give you a "plan as a prompt" using the Max thinking, while telling it to make sure to conserve tokens not doing useless validation. but yes it does run out if you make it do the 3D models, just make the game with placeholders then use a separate session for models
>>110131463I think that largely depends on what you're trying to have it model. I'm working on a PSX style game and modeling that stuff barely even scratches my 5 hour window.
>>110131489>>110131515That's good to know.I was thinking of just making something small to start. Maybe equivalent to about 2 or 3 worlds of Mario 64 in size for example (e.g. Bob-omb Battlefield, Whomps Fortress, Cool, Cool Mountain). I can do most of the general programming and mechanics and UI but I struggle a fair bit with 3D modelling and terrain and also shaders.
>>110131556anything related to assets like models and shaders, should be kept in a separate session, due to their high probability of derailing everything if a simple mistake happened. goodluck slopping anon
>>110130722You will never be a real developerYou have no skill, you have no creativity
>>110131613go back to /gedg/ and /agdg/ this is our thread now luddite
>>110131613no slop for you!
Cris do you ever actually realize that you won't make a good game or are you just delusional? What's your honest thoughts on the fact you've spent so much time "learning" and haven't actually made anything besides baby's first tests
>>110130722>AI Game Development General #1can just go to /agdg/ for a faster version of that, though.
The diference between these is 1 years.https://www.youtube.com/watch?v=MY1RDpLk2KQ [Embed]https://www.youtube.com/watch?v=ZK7iUPLnt4Y [Embed]https://www.youtube.com/watch?v=gH9kcx-04lI&t=8s [Embed]https://www.youtube.com/watch?v=4Eo9cX0yaBw [Embed]https://www.youtube.com/watch?v=31rBSiJwl9E [Embed]
>>110131764we don't need annoying luddites here, go back
any diteys
You know what, yeah, maybe a real AI thread is going to be better than dealing with the schizos, hopefully the moderation on /g/ is better. >>110130830You will be my anchor now, congrats.So, I had already added most of the dragons (I still need to work on the models, and the balance is all over the place), but I also added ground units; which will serve as mostly mana farmers for the player and will just proceed to fuck shit up from the ground: lower firepower than dragons but last much longer and won't cost as much. But then I was like 'yo what if I get autistic about it and create a different version of a guard for each element' and now I have 5 different guards that do different things, which is going to be a nightmare to balance, but yay!
>>110131818this is very great anon, what model did you use? also how did you make those hexes and animate them?
Anyone decompiling old games? God, it takes fucking forever to translate and binary verify on a local model. At least I managed to find a correct compiler for that
>>110131903i did use a jailbroken kimi to decompile some old flash games and 3ds games, too intimidated to run huge local models atm. what are you working on btw anon
>>110131818This looks awesome.t. anon scrolling from page 0
>>110131925Just testing if it's technically possible to fully decompile some early 3D games from 00's using mostly local models, to save on tokens. I don't have high hopes in it, but looking at functions table slowly filling in with translated functions is mesmerizing on it's own.
>>110132021a guy used qwen 3.8 27b at q6 on two 3090s to recompile fable 2 from xbox 360 for pc its on github
>>110131852I design most of it, but I didn't draw anything no, for this game it's really just claude, it's on unity and I can't trust astra with it, she's fucking retarded, well, she helps with creating 2D sheets for the models, and then opus handle the creation and improvements of said models. >>110131950Thank you nonny, it'll take a while before it's cooked tho, right now it's basically just a tech demo and me trying to figure out how far I can push the engine while giving the player muh anime power fantasy, and keeping it fun (well I'm sure some people would love to just have shit to blow up and spells to do that, but I want something a bit more involved than that).
>>110131903I feel this approach of "byte-exact match" is so retarded. If you’re exactly reproducing the bytes, you might as well just copy them?!?!? -_-
>>110132084You are not reproducing bytes, you are recovering C shapes that produce the exact same assembly binary as original game, on the same compiler, with the same build flags. That way you are making sure that it's same C code that devs used, and not a hallucinated vibeslop that will need further fixing. Even if you are planning to rewrite the shit out of it, you still need a "gold standard" base game code that's guaranteed to work exactly as original, to use it as a reference.
>>110132188>That way you are making sure that it's same C code that devs usedYou can't get the same C code the devs used you idiot, compliation is a one-way process that destroys information
>>110132452Learn to read, retard. And yes, you literally can.
>>110132475No you can'tYou can write two different C functions that compile to the same assembly code - you will never know which one the developers used
>>110132485>You can write two different C functions that compile to the same assembly codeProve it (You can't).
I'm not currently paying for any models, just using a free stealth model (which is likely some version of GLM).I also haven't tried any 3d mesh generation, only using freely available assets for now.I'm currently just messing around with creating different plugins for basic game functionality. My goal for now is to just implement boilerplate functionality that can be used across different game projects that I come up with. Will post a webm if anyone wants to see.The LLM did a good job on the camera plugin, but so far hasn't done the best work on the 3d player movement physics. There are still issues to work out, and I'm not convinced yet that it would be faster to keep trial-and-error prompting the LLM rather than just fixing the code myself.
>>110132511Are you new to programming?Dividing an int by 2 can be replaced with a bitshift
>>110132592Yes, and? The devs could also have effective noop in their function, that's skipped by compiler and it won't be recovered. It's functionally exactly the same, that's what matters.
>>110132616>It's functionally exactly the samenow you're backpedallingyou said "The same C code"you will get code that produces the same output, you won't get the same C code, if they compiled with optimizations on it will be significantly differentobviously you dont get names and comments too
>>110132630Code formatting and naming is irrelevant. Tiny changes in function shape can result in completely different binary. Forcing decompiler to iterate until it matches gives you the closest function shape you can possibly get. Also some games were literally shipped with debug symbols by accident lol, so there's that too.
>>110132676>Code formatting and naming is irrelevant.If you want to work with the code it's very relevantIf you compile with optimizations on then functions get inlined, you dont even know where function boundaries are
>>110131613>You will never be a real developerI don't want to be one>You have no skill, you have no creativitySeethe all you want in your dead thread
While I can 3D model and code on my own, I'll admit after seeing a video using Claude (I think) modding GTA San Andreas it got me thinking about how hard it would be to make something similar to a 3D era GTA game but use the saved detail on models and builds on interiors, vehicle physics, and generally making the world feel more lived in with a fairly small city. I'm not sure how far along this tech is yet, but from what I've seen looks promising if only to offset me being a single developer and being able to fill in blanks between key assets. Any idea what the costs associated with Claude for doing that and how many tokens I'd get/use per if going for a vague tricount between mid-to-late PS1lN64 and early-to-late PS2/Gamecube era 3D models depending on the assets? I'm already working at basic skill limit for models right now which I'm happy with after years of modding older games, it looks fine in P:Z and GTA so I figure it could be a gamechanger if only for a personal demo/proof of concept.Sorry if I repeated myself a bunch, on low sleep from having to deal with PHP related bullshit for remote work.
>>110130722I should mention that /v/ has some pretty neat Claude mod threads here and there and should be a good starting point if you want some inspiration
>>110132687Okay, I can't deny that there are some losses. But it still converges on a single result, instead of making wild guesses based on ghidra disassembly alone.
>>110132712I'd wager you can do something like GTA 3 in two months time, assuming you keep the exact game design, and you don't do anything crazy with missions or story wise.Most of the time will be spent on iterating on gameplay to get it on par to GTA and/or for you to figure out how to design any new systems you'd want to add in, and also tardwrangling it to get better models and animations: it's not great at either yet, it's 'serviceable' but not great. You can probably get a basic demo going in 2 days at most tho desu, but the AI will be trash, I'm thinking police behavior will require a fair amount of testing too, etc. A game the size of san andreas: basically the same, the code and functional side of things; claude is going to completely destroy. Everything else: not so much, it'll be up to you to do a lot of the work. The 'vibe', the art direction, trying to make vehicles that are legally distinct while having a fair enough amount for variety, and then of course voice acting, missions, etc etc. Claude is also not great with sounds yet, serviceable, and sometime he'll get something right right from the get go, but sometimes you'll have to tardwrangle it for two hours to get something you'd otherwise think to be rather simple to get.
>>110132712I think you could get quite a lot of assets out of Claude for fairly few tokens. You'll have to provide some decent directions and textures, though.It's only like $20 to find out.>>110132782>something like GTA 3 in two monthsOnly if you're unemployed.
>>110132883>Only if you're unemployed.Yeah no shit, what did you think I meant?You think you're going to make an open world game the likes of one of the biggest franchises in the world while working 50h a week at mcdonalds?I also meant 2 months of using a 200$ sub, don't even think of tackling something the size of GTA 3 (let alone SA) with a 20$ sub.
>>110132712Not sure to be honest. I will think it will be hard because no matter how much Claude gets done you still have to make everything else or at lest define the game design, we are talking about, hundreds of pages of mission designs, story progression, characters, ideas for missions, how the world is, how the world is going to look like, what period you want to tackle and so on. I think that you will spend most of your human time spend writing documentations files to help Claude keep track of things and don't get lost in the process. The others anons have the right idea. It will be a lot of work but you can get something out of the door in a couple of days if you don't really care about anything but a simple playable sandbox. >. Any idea what the costs associated with Claude for doing thatThis will be your main bottleneck, even if with the 200$ sub you are going to be hitting that limit a LOT I'd wager. If you don't use the API and only stick with your sub allowance I think it's going to take you like.... 1 year at least. To make a fully playable early GTA clone working on all platforms with all the major quirks and bugs fixed. And I'm thinking you aren't going to make your own game engine for this and instead use something that is already battle proven like Godot or unity or whateverI would say, since you can 3D model, let Claude take care of the code and do the 3D in the meantime
>>110132922>You think you're going to make an open world game the likes of one of the biggest franchises in the world while working 50h a week at mcdonalds?Clearly not. Which is why I specifically pointed that out in your response to some guy not actually having used Claude yet and complaining about the job he has taking up a lot of time.
>>110132928Conceptually the idea was more or less using it to help me prototype something quicker than otherwise and then slowly replace what I can or optimize any mods it puts out, while making the key models and sounds myself, coding what I can here and there. I figure it's a decent enough use case similar to how I use AI when I'm out of ideas for poses and gen a batch and then use those like I would any reference photos and draw the art myself. Which is to say, I have a varied skillset enough to make a game, but right now I'm worried about how long it would take just to prove to myself the game concept isn't entirely a dumb idea. While $200 a month is high, it's not overly high and I could swing it if it meant being able to get at least a working prototype to test, tweak, and eventually send to some friends for feedback on the direction. As far as writing, I've written probably an excess of design documents over the years but I've been stuck in a holding pattern of assuming the idea is too complex, paring it down, then doing that and nauseum without testing if it would work, or when I do it works but I get over confident. Either way, I figure this might be a decent time saving tool if only to prove to myself the concept isn't totally trash before fully committing, and $200/m isn't great but it's also not horrendous. Any ideas for what tokens would be if I downgraded to late N64/PS1 and PSP tier tricounts on assets as they are on average lower than PS2 by about ~1000-2500 and smaller texture sheets.Also I figured I'd post the video that had me thinking about this since last month between 3D modelling sessions and writing sessions. >https://youtu.be/isjBe4khZZcI'm gonna be perfectly honest, I'm a huge car guy and I kinda want an excuse to create entire fairly low to mid poly car families with realistic options and it seems a game like this might work well as an excuse for that if it goes well. I'd be fine just making the prototype basically a sandbox.
idk how to feel about the fact this general has genuine retards arguing about the topic, and filters dramafags and namefags. im not even an AI slopper
Triple A Graphics in the browser by next year that outperforms Unreal.
>>110133053Pro tip: Use Opus or whatever the latest model is called only as a master agent, and have it call sub-agents with cheaper or free models.
>>110133058There's no actual drama from real people, it's all falseflagging and then autists falling for the troll and getting mad over it.If the retards in agdg wanted to make actual video games, they'd be working on making video games or at the very least discuss the topic, the only people that do both are fags using AI; that's all there is to it.Now fuck off back to your thread before you attract them here.>>110133053I'm doing very low poly (ps1-early ps2 era) models in my games, my infantry girls for example (including hair, clothes and guns) are <2000 triangles. You do not count in 'tokens', this is retarded, because you do not want to do this with api tokens; this will costs you thousands of dollars to do anything like this.Instead, you can just try with a 20$ claude sub, and you see for yourself what you can make: try to stick to sonnet as much as you can; sonnet will eat up most of the gamedev logic work very easily, opus is overkill for 80% of the systems in gta3 (if we stick to using it as example); only use opus for things like physics and then heavy refactors or optimisation work (you'll want to use the big brain AI to decide what goes on each threads etc). And then you can easily figure out how much you really need, considering the 200$ is not x10 the amount of tokens as the 20, but x20 (that's how the kikes get you, so that you don't get the 100$ sub or you'll feel like you're losing on money). With a single 200$ claude sub, I can have 3 terminals working on my three different games at all time during the day on opus 5.5 xhigh, and I couldn't even reach my 100% quota.
Guess the Engine. I'll give you a hint: This is raytracing and nanite.
>>110133152listen bro; we're very hype about it; but that's all these is to it for now, we get it, it's cool, the future is cool, but stop it, we're waiting here, post results when there's updates, we don't need the constant reminders of what everyone already know
>>110133053PSX style assets are very cheap in terms of tokens. Don't know the exact average token amounts, but having Claude create 44 lab props for a level cost me like 5% of my 5 hour window. They would probably fall well below what most people could stomach in terms of fidelity for such a type of game, though.I'd honesty recommend starting out with just a $20 sub. You'll soon discover, if you actually need more once you started and you can just upgrade. It works perfectly for me, as I only spend like three hours for a few days a week on the project.
>>110133152go back to agdg>>110133167>AI chad defending the comfy thread
>>110133053Since you have written the design docs, why not feed that into Claude and ask him to come up with a plan to make a very simple playable prototype. Especially since, you want an excuse to play around with cars in a sandbox . But is GTA really the game for that? Why don't you try something adjacent to Driver series or something along those lines, not only your scope is going to decrease but you might get a chance to play around with your idea before really trying to commit if you want to also add and mod GTA elements to your game, especially since you want to pivot you can do that with Claude.
>>110133182It's three.js. The number one vibe coding renderer.
I'll respond to other helpful posts here after I have a shower and make breakfast>>110133197Funny enough, Driver 1 and 2 were actually what spurred the idea originally but it ballooned to a point where I got discouraged before cutting it back down to something I could make. After replaying Vice City and spending time between missions examining models and otherwise just waffling around the world doing analysis I figured maybe I could combine the two in a sense. I am more keen on attempting to use AI to prototype this among a few other highly experiments ideas for similar reasons to what I assume you are pointing out here such as third person on foot movement being somewhat redundant given my interests in the concept, though Mizurna Falls was further inspiration. It's just too hard for me to say right now. Like you say, the thought was to prototype it roughly this way and do some key assets where needed and if it actually feels like a viable direction I'll pursue it further, or pivot.
>>110133167I just wanna talk about 3js
>>110133239Anon is right, and it's smarter anyways because the 3rd person part is not what gta is about, 80% of gta is about driving around, you want those to be good and feel good, it's a good first focus to have.And honestly, claude probably would've proposed you to do this as a first slice anyways. >>110133252Well I don't mind discussing it; but what's there to discuss that we haven't discusses yet?Tell me when I can drop godot and use threejs for my games, I'll jump on it as soon as it's ready; but it's not, so let them cook.
>>110133252sorry anon, mistook it for agdg (you) farming. you may talk about 3js
>>110133284>>110133239Also, by the way, you might have a better chance telling him to use san andreas as the inspiration, rather than GTA3, merely due to the fact that SA has an autistic fanbase and the game has been completely retro engineered over the years in random russian forums and for SA:MP etc etc, so the clanker will have a much easier time finding info on SA than it could on GTA 3.
>>110133288I'm actually from webdev general and do networking stuff. What I'm actually experimenting with atm isn't gamedev but networking and versioning. As >>110133284 said I'm pretty much just waiting for the three.js guys to cook as far as the renderer goes. Creating anything we want is already a solved problem with AI.So what I'm looking at is a version manager + downloading system. Basically, no one wants to redownload a game every time they play it. That'd be retarded. So the idea is a game hosting site where if a user likes the game they're playing they can click download it it downloads to OPFS. Then next time they come back to the URL if anything is updated, it only updates the stuff that's changed instead of redownloading the game, like any other game outside the browser does. I'm researching this library for it atm.https://kaset.dev/concepts/versioning.html
>>110133434Don't you think the prices of maintaining this are going to implode as storage prices keep rising?Also, this isn't going to be affecting most; but I'd never use this system simply due to the fact that I love modding, and this would make modding harder (not only old school modding, but even more so now using claude to do whatever I want, how do I achieve this if it's not on my pc? I guess a symlink bullshit system, but that's just going into frankenstein territory, idk)
>>110133470 What do you mean? Think of it like this>Store game on github pages, no different than storing game on steam>User opens it, it downloads in real time, user plays it>button pops up saying "Like this game? Download it to system" and it saves to OPFS>the library looks at github so any time there's git diffs, when the user comes back they're only downloaded updated file chunks>The game itself still lives on the user's system and is moddableIt's like the browser is becoming Steam Client essentially, the games are still saved locally, and the devs can just push updates and it's handled.
>>110133515sounds like a security nightmare
>>110133525How so?
>>110133515Well that's my point bro; you're not steam, and github isn't going to allow you to store petabytes of game data for free. But I don't hate the idea; if the execution is good, but we'll see. I like the idea of being able to allow people to try my demos on the web tho, for example, this is so much easier.
>>110131903I am. I'm skipping the `byte exact` approach and just having DS4 / Qwen3.8 flash use back-substitution / ghidra / pcsx2 mcps / a shitload of auto generated python scripts to generate C code that is as human readable as possible. From there you can manually craft an engine around the behavior of the original game. It works but yes compared to Claude its much slower.>>110132452"pure" decomps work by finding the exact same compiler version that the original devs used and using the asm to generate C that when compiled produces the exact same output. It's possible but its really tedious and hard, made less so with ai automation.>>110132592This is pedantic, it doesn't matter if its literally the exact same C code character for character, only if each function behaves identically
>>110133557>for example, this is so much easier.NGL I'm kind of hoping Spawn.co picks it up and picks up the renderer. That'd be the golden age of vibe web gaming. Just creating triple a tiers in the browser.
>>110133560>only if each function behaves identicallyOptimizing compilers inline functions
yes, yes, good job clanker-kun, you've decomposed the armor of the design I want you to work on.....
>>110133126Hey, I remember you from /v/ thread. Your sound design is top notch. How did you make it?
HOWEVER....
>>110133618It needs to decompile it first so it can understand it.
>>110133590look at the sm64 decomp if you think this is impossible to do
>>110133638I didn't say it was impossible to decompile games, just that what you're getting out is not what was put in
>>110133622The only tip I can give you is https://github.com/Leonxlnx/earshotThen it's just a matter of tardwrangling it and trying to find the magic combination of words that'll make it do the sound you want, sadly, it's not quite 'smart' enough to understand everything (for example, telling him 'I want it to sound like X' is only going to lead the clanker to try to find a spectrogram of that sound somewhere on the internet; chances are there's none). There's sadly no real secrets, I haven't tried feeding him actual sounds that I would record myself and tell him 'use that as an inspiration and make it more metallic' or whatever, maybe this is the right way to do it. Also, humble bundle software section; they often often sound banks of hundreds/thousand different SFX, I'm thinking using those to tell him to use as inspiration would be good (it's just a huge pain in the ass to download them and set it up).
Now that rendering, performance, and coding is about to be solved; What's the solution to assets boys?
>>110133746no idea still, all the models i used tend to shit the bed with assets and create basic ass stuff, if not broken stuff. but i suppose it would need something that's "multi-step" to work?
>>110133746I'm a graphics programmer and the code Claude produces is quite bad unless you actively shape it into doing it properly. E.g you need to know what is efficient and what isn't. If you just let it vibe it will do shit like 90000 draw calls when 9000 would suffice and not optimize or cull anything. It tries to get to a working state as fast as possible and stuff like that falls by the wayside.
>>110133805got any tips anon..?
Alright boys, gotta take some labs I'll chat later. But a few replies before I go.>>110133805Yeah I heard Claude is bad at visuals still unless it's physically ripping stuff off. I was hoping someone would suggest something else. At most I'm thinking a low poly mesh generator at the very least so we have a starting point. Fireship mentioned a generator that went from shit to generating a fully functional Rolex watch. I'd have to look for it but I'm out of time.>>110133777>multi-stepHave you played with Spawn yet? It's multi-step and works pretty well, given it's low poly. They did a stress test earlier this month where people were creating shit in real time on the same server.https://x.com/jsnnsa/status/2105568049428607411
>>110133679wtf, I didn't know about this, thanks!
>>110133805>>110133777Okay I really shouldn't be wasting time on this mid lab but here's the video on mesh generating I mentioned. It was Asta.https://youtu.be/2Xiljy4xzbc
>>110133830Well the best thing obviously would be to have computer graphics programming experience so you know how to optimize, but in lieu of that you can always ask Claude to work in *very* small steps so you can review it, and at each point ask if the chosen rendering implementation is as optimized as possible. In my experience every time I do this claude will go "you're right I did the lazy thing because I'm a bitch you actually want to do X" and you effectively have to second guess every single thing it generates, one of the worst parts of working with ityou want to ask for things like "are we using frustum / occlusion culling?" "is this cache efficent" is "is there a way to make better use of simd here" "could we use a compute shader to optimize this" etc
>>110133914>"is this cache efficent" is "is there a way to make better use of simd here"you sound like a performance larper
>>110133914What's your take on KTX2?
>>110133926what does that even mean, those are relevant questions to ask when you're trying to optimize shaders. Using Tracy (free profiler) is also a really good idea, because then your agent can just read the profiling information directly instead of having to guess
>>110133937>those are relevant questions to ask when you're trying to optimize shadersno they aren't, you have no control over SIMD in shaders
>>110133929It's good, texture compression in general is worth it for bandwidth and vram savings, use it.
>>110133914thanks anon, noted
>>110133942yes you do, look into subgroup ballots they added in vulkan / glsl and even if you couldn't control it directly being aware that GPUs leverage SIMD changes how you structure your actual data layout / gl_WorkGroup size to be most efficient
>>110133946No downsides then? The Youtuber I learned three.js from made a ktx2 conversion tool for the web and it seems too good to be true.
>>110133968It's texture compression, so the downside is that it's lossy technically but in practice nobody is ever going to notice it.
>>110133968no texture compression isn't too good to be true
>>110133983I was worried about what >>110133976 said about it being lossy, primarily. It seems like how efficient the compression was is too good to be true and there'd be some catch I didn't know about.
>>110133994it's good for pretty much anything that isn't text / pixel art, it's not a jpeg or anything
>>110134025tf did you do anon to require that many continues, edit a whole ass movie script?
>>110134025What are you doing?
>>110134002Alright. Noted. You have to understand when you see numbers like this, it always seems like "What's the catch?"https://x.com/iced_coffee_dev/status/2066567470668718562
>>110134070GPU compressed textures tend to be about 4x smaller in my experience
>>110134070>>110133152Web really is going to become the most efficient game technology, isn't it?
>>110134110Already is, given the insane amount of games you play on your web right now
>>110134110texture compression is not "web"
>>110131818How'd you do the book? I want to do something like that where it seemlessly flips to a new page while text sticks to it but I'm unsure how to do it beyond a general idea
>>110134070im not a webdev but like, how tf does a 39MB file take 342MB in memory?? shouldnt it also be 39MB? genuinely asking
>>110134152compressed for storage, decompressed for use
>>110134152>Why is file inside .rar not the same as file outside .rar?
how do you handle multiple agents doing multiple tasks
>>110134110It's not, you can do all this with C++ / Rust and be way more efficient than webshit will ever be. >>110134133Render the text to the page. (The page is a two triangles, uv map the text to the page, animate the page flipping and the text will move with it)>>110134152JPEGs are compressed, when you load it into the GPU you have to decompress it. GPU compressed textures exist to avoid that.
>>110134166in the same project, i mean
>>110134158>>110134164but the image said "No Compression, 39MB file, 342MB in memory"
>>110131903>>110131925Advice on decompiling flash games and recompiling? There is one I want to play but it doesn't play beyond a certain point. I know what needs to be changed but I'd have to access the code and edit it so it runs as desired.
>>110134133I literally told it to imagine a book opening and pages flipping like that. Though it was a bit tricky, it's a rather complex animation and complex task, it got it wrong a couple times (not horribly wrong, but wrong still) where it for example didn't add extra 'cosmetic' pages (which you need on each sides, obviously) and this lead to weird interactions with the cover, and at first the cover wasn't a single piece either.Basically, ask it to first put into words (that way it'll use a vocabulary it understands itself) how he would depict an animation of a book opening and pages flipping, and what you'd want out of it (in this case, I wanted a grimoire style-UI where the left page was a specific something and the right page another something, but at first I had a manual page flipping system (which was horrible UX)).You'll just have to iterate until you get something you like, I'm not done either, it's not perfect yet.
>>110134197The imagine is clearly retarded bullshit
>>110132059Not either of those anons but how did you do the 3D?
>>110134181Friendly reminder to the rustrannies who think they can speak to us C++ and Javascript Chads like we're equals. Everything you make is shit.
>>110134201What is the end goal with this one?
>>110134069continuing
>>110134198>>110131925so far i tried decompiling/recompiling flash games, i could freely edit the strings as well as sprites using claude sonnet. however, i couldnt touch the UI placement and some scripts/codes like scene transitions and such. i wonder if an anon here knows how to fully edit a flash swf file.
>>110134181>>110134201That should be useful, thanks both of you. Also did you get it to model it as well!?
>>110134215havent seen anons arguing about programming languages in dev threads for a while, nostalgia
>>110134264thats how you know the thread got autists instead of namefags and drama (you) farmers
>>110134212I prayed it would get it right and I called it a retarded clanker a few times and iterated. Though the broom, the gun and the enemy models here are just stupid placeholders from the unity story, and the player character is literally unity chan as well.The dragons, buildings, world, clouds, etc that's all 100% clanker made, no blender or anything from me (well, yet, I'm in the process of fixing some models in blender right now). Opus (astra is still shit at most work) is pretty good at most 3D that's not organic honestly, it'll do a pretty okay job making a model for a weapon or buildings etc (here they're basic tho, I didn't tell it to spend any time on it because I'm just testing things, so it's not quite a good idea of what opus can do). It only fails miserably when trying to do anything that's living and should move in more 'natural' ways. It does get some things right every now and then, pic rel for example; the 'earth' (force) drake; big boy that's going to bounce around and is mainly there for material destruction. >>110134237Anime.Literally 'what if tanya the evil in a video game' but I also wanted to go full on gurren lagann bullshit, I want the world to break, I wanna dig holes through the clouds, I want to make the moon explode. This is the end goal, but I do have plans for gameplay too, but those come later, now i'm still just building all the infrastructure etc (all of it is clanker made, there wasn't any updated and good enough library for all that I wanted to do with the clouds or the geo destruction, so opus has to do it all from the ground up). >>110134242See above.
How do you stop yourself from coming up with new ideas and instead actually finish one thing
>>110133867Decided to watch some Astra videos after rewatching that. Is it actually this good? Has anyone used it? Seems they're using it with Blender.https://youtu.be/I9Oc5AwAxBwAlso does anyone else feel like when you post a product asking if it's good, you get social anxiety that someone is going to accuse you of trying to sell the product? kek.
>>110134315Sell me on Halo-definitive
>>110134318It's good but you need to weight that against the scope of what your project needs and how much money are you willing to spend to get it done.
>>110134337I mean, however much you spend is less than paying a team of graphic designers right?
>>110133746Easy. Just make some retro inspired game, that doesn't require high fidelity assets.
>>110134324nothing exciting, my own implementation of OpenCE with the goal of demaking the campaigns of 2 and 3 in it.
>>110134343yeah but that doesn't mean you have hundreds of disposable income every month for tokens, does it
>>110134374sounds good BUT IS THAT BIONICLE
>>110134318This is pure BS, astra is absolutely trash at 3D still, even in ultra. Pic rel; design vs actual blender result.
>>110134372To me personally it seems like Cel Shading is the way to go. People love the anime games look, but they're typically not super HD models, the shaders do all the lifting.
>>110134402Good thing you can get the shaders most anime games uses because it's available on github
>>110133746Procedurally generated art.I need to take an art direction course.
>>110134416Oh is that so? Source? I watched a video of a guy reverse engineering how Guilty Gear Strive creates such good Cel Shading a while back and it was interesting. They're personally coloring in all the shadows by hand instead of actually relying on the shader, without actually caring about how lighting looks, tailoring shadows to how good it looks with the model.
>>110134389Yeah, been trying to brainstorm what a good bionicle game would be like, haven't gotten that far in itopen world rpg? linear action? idk
>>110134402Yes, but you still need decent models for that, that really work with the style. I like to avoid all but the most basic 3d models for now. With 2d enemies I can just slop them out with some local image model and animate them a little.
any anons here working on like an H game or something of the sort? I wanted to ask, but what image generation model do you use or would use?
>>110134559Novel AI is probably the best, the rest doesn't come close.
>>110134559When it comes to anime porn, nothing beats local models. It's either illustrious or anima variants these days. And the amount of various loras and tools people made for those is crazy. Any smut in any style mix you want. Can get as consistent as you need if you train a lora yourself.
>>110134559>but what image generation model do you use or would use?Definitively NAI for online models or if you want local you try Anima both can gen nice images but if you want to make an anime game (SFW parts) at least you should use Higgsfield which seem to make really amazing anime so far.I think I saw a lot of anons over at /v/ talking about how Claude doesn't care too much about the contents as long as you trick it a bit. Heard some anons modding Artificial Academy 2 and creating mods for it but not really sure if everyone is going to get by the initial filter
>>110133126cool ass game. disable hitmarkers thought that shit is literal cancer
>>110134526What models would you suggest? I was going to look at Astra but >>110134399made me change my mind kek
never imagined the most comfy and civilized dev thread would be the AI slop one. im not slopping nor using AI, but it's nice seeing people actually posting progress and helping eachother with concepts and stuff.
>>110134695>turns out the creative people that just want to use the cool new technology to make shit they've wanted to do for years is... not filled with retarded third worlders whining about the end of the world ad vitama true mystery this one is>>110134659Yeah there'll be an option for it, I wanted it in because I'll have very very heavy fog and so... it helps sometimes, it's very GTFO inspired. I'm very very far from done, though.
>>110134695it's easy to post progress when you don't have to do any work
>>110134713>ot filled with retarded third worlders whining about the end of the world ad vitamlel'd the only ones whining are the murricans >>110134698>turns out third worlders also like to use AI to build cool shit
>>110134725not really the progress thing, meant theres no basement dwelling trolls thriving on izzat and engagement farming like the rest of the generals.>>110134743>americans still are at work or asleepoh well, until the muricans open the sitenvm americans woke up >>110134777
>>110134559It depends on the game you want to make, I think that if you are going for weg games, you are in so much luck to be honest. but if you are making an actual H-game you aren't that lucky. As the others anons have said, Illustrious and Anima can do the heavy lifting when it comes to Art, you can keep characters pretty consistent right now with a mix that you enjoy. But sadly local models breakdown more often due to increasing complexity of the poses but hey, you can play a lot of local models and if you haven't I would suggest to check out the generation thread here.What game do you have in mind, anon?
>>110134637If you go with Anima, use QWEN2.1 to remove the background of character images.>>110134671I am using Claude, but I'm probably the wrong person to ask for decent models, as they are all pretty PSX in terms of fidelity.
>>110134845A good renderer and good textures can make meh models look good. I'd imagine we could do image gen for the textures anyways. When you look at the renderer I keep hyping up, rasterized side doesn't even look all that impressive. The detail in the armor looks fine but without the textures and renderer it looks ps2 tier, like something you'd see in Final Fantasy 12.I think low poly models could be hardcarried desu
>>110134918>image gen for the texturesI'm already doing that, but it's sometimes tough to get a good texture out of the models I'm using. Especially seamless textures aren't all that easy.
>>110134987There's good texture sites for most stuff as far as materials go, but I don't know any for your use case of low graphics.
>>110134571ah novel ai... I heard it has improved since last I used it which was back in 2023>>110134637yeah local sounds nice because obviously who wants to pay for a subscription, though don't they require a beefy PC? And loras are basically the art style, correct?>>110134638I'll have to try out for different things, its just that with NAI keeping the consistency of a character might be tricky>>110134807Illustrious and Anima, neat! But as I said before, don't they need you to have a beefy pc? Like I imagine it would take minutes just for one pic. As for what I have in mind, there used to be a game called Saria Reclaimed before the dev sperged out and fucked it all up. I'm thinking of something like it with its own twist and stuff. But im just in the planning phase lol
>>110135035>its just that with NAI keeping the consistency of a character might be trickyIf you haven't used nai in a long time you might be surprised to learn they now offer a "Character reference sheet" mode where you can upload your own references and nai will be able to keep the model consistency locked in. You should go around AI art slop threads here on 4chan or around twitter to see what some people are doing with it
>>110130722so about that 'footfag in a shoe shop simulator' project... ?
>>110135146Please make it. It's gonna be a gold mine
How much are people getting paid to shill NAI?
>>110135035>, there used to be a game called Saria Reclaimed before the dev sperged out and fucked it all upIs it a RPGM game?
>>110130722It's not exactly game dev, but I can drop here the progress I'm making on openuma as well. Just finished the first career loop through the URA Finale scenario.
>>110135066good idea, it just that here I have gotten actual info and responses lolOn twitter those "artists" are too secretive on how they create their gens. Though hearing that new feature of novel ai sounds promising. I guess they forgot about text gen?>>110135173Yeah if I remember right it started off as RPG maker and then turned into a shitty vn lol
I'm making a gsg with EU5 as a reference (except not shit). >writing code 24/7 in a container which automatically pauses and starts whenever my limit is reached>i get notifications whenever it has something for me to do>sonnet writes code and orchestrates agents>opus reviews and integrates>gitlab CI which cucks sonnet whenever it exceeds a performance target>everything written in safe rust with modability in mind>engine is barebones, the whole game is basically written as a 'mod' so that anyone can do anything with itWork is slow (for an AI) because it's a lot, and I only have a Pro sub. It's on milestone 3 out of 14 after 1-2 weeks. Still not sure what to do about art and music since using AI for that just feels soulless. Anyone have anything they'd like to see in this game?>inb4 use albanians as fuel
>>110130722/aigdg/ - AI Game Development GeneralEDITION>"Just use AI, nigga" gamejamsNext: https://itch.io/jam/agdg-juan-jam-2Previous: https://itch.io/jam/agdg-juan-jam>Agentsopencodeclaudeantigravitycodex>How to startNewbie? Ask AI.Already a programmer? Just act as a director. Use software engineering models (Waterfall, AGILE, scrum, etc)>How to WebMhttps://obsproject.comhttps://github.com/argorar/WebMConverter (works via wine)>Linkshttps://itch.io/games/tag-aigdg>Friends>>>/vg/agdg/>>>/g/gedg/>>>/g/vcg/>Previous>>123>---x---x---x-----x-----x------x----x------xShould our jams be agdg-like where we post our main game progress or gamejam-like where we create a game for each jam
>>110135219Pic related. I'm just sipping rum and coke while my AI sweatshop works it out. I might have another playtest tonight if it finishes M3 before I go to bed.
I have been recomping Shandalar to rust since the regular version kept crashing and corrupting my saves
I told ChatGPT to port a Unity marching cubes compute shader to Bevy and it did, the mad lad.
>>110135231I like this OP but maybe a couple of better wording? Or maybe we could start to compile a rentry with all the good stuff like how to decompile games, recomps, how to extract assets and so on
Another new Three.js renderer that looks great just dropped. The guy making the other path tracing renderer replied to it. Open source and part of his web starter kit, so I assume it's a vibecoding starter pack.God I love browser Chads. We eating good boys. I'll go through it and see if it's worth using.https://x.com/PashaIGooD/status/2113274755830992935
>>110135321Just an idea, feel free to improve it.
>>110135332Looked into it. Looks like he's making a full on vibecoding engine for people rather than a renderer like the other guy. That's cool for pumping out games fast but I'll wait for the renderer personally.
>>110135035Local anime models are notoriously tiny, so you'd be more limited by the speed of generation. Loras are small chunks of neural network slapped on top of the model. They can do anything you trained it for, from styles and characters to concepts and skills like extracting a depth map or making ultrawide shots. Also tools like controlnet allow you to transfer poses from a skeleton to your gen. Basically the main selling point of it is that's it's very flexible because of community around it.
>>110135318What's to port there? Isn't that a well known and documented trick? Might as well just asked it to implement it from scratch.
>>110135035I can generate 1024x art in 50-60s with my RX 7600, 8GB VRAM, on Windows. It's a bit faster on Linux
>>110135524I found the Unity implementation particularly elegant and it's useful to have something to compare the output to.
>>110135525Surprised an AMD cars can do this. What are you using for local diffusion? I have a 9070XT so it should be much faster for me.
>>110130722i am an AI chad but i shant post my 'gress in a tranime threadif you want us to leave agdg you have to make a pleasing OP to us the AI daggers
>>110135657NoobAI-XL, but Illustrious-XL works too. LoRAs don't hurt my runtime, but detailers do. You should be able to do much better with a 9070 XT, yes.>Surprised an AMD cars can do thisWe have come a long way, brother. 3 or so years ago AMD couldn't generate anything but tears from their owners.
Are you guys slopping up the art or just code? I actually wanted to learn the art side of things. Making textures, 3d modeling etc
>>110135821I can slop 3d models with Sol 6.1 and Sonnet/Opus, they're pretty decent at it for stuff like furniture or grayboxing a map
>>110135816So I setup comfyui and then get all the other shit from huggingface or whatever?
some progress on the UI
>>110135830Yeah. Ask your preferred AI for help setting it up, Comfy is a confusing bitch
>>110135838Thanks fren. I'll check it out tomorrow.
>>110135830You get the models from civit. Sort by popularity and model type and pick a mix you like.
Are any of you scared of someone cloning your games?
>>110135856Yes. I won't show any more of my game once it's actually in a playable state.
>>110135856its the era of abundance, more good things.
Just curious as I may want to add more AI usage for gamedev, currently I just use free AI and ask it questions and then usually use the knowledge to build out the code myself. Nothing too complex, the last time I tried complex it only partly worked with some of the code just not working as intended. Spent a few days trying to get it fixed yet the AI kept failing, had to figure it out without it.Anyways what AI's do you guys use and in what capacity? What kinda game? I assume the kind of game heavily affects which AI is best for it.
>>110135821I let Claude slop some 3d models, and try to create textures and sprites with Krea2.>>110135856No. It's just a hobby project and probably not something anyone wants to copy.
>>110135856Not really, they can clone all they want to be honest. I don't care because I still want to create even if they are going to clone it
>>110135914Claude. I haven't written code in months
>>110135821art is what ai is worst at and will always be what requires the most taste. I'd view the ideal AI workflow to be even if you let AI do most of it you using art side stuff to set up what is normal and have it build in line with that would still result in something that looks much better
>>110135914No, I used to believe that, but currently (it changes every now and then), opus is the best at literally everything save for image generation (then it's gpt). Mind you, this means that gpt is 'ok' at making textures, but opus is just as good as well so, unless you need design (which does help), then you can just get a cheap 8-20 gpt sub and then go all in on claude and you're better off than any other options. Fable 5.5 is coming soon too, maybe astra 6.1/6.2 will be better, but currently astra is just much more expensive while being 50% more retarded.
>>110135914It used to be like that but with Opus 5.5. Things really have stepped up a notch here. You can do pretty much everything you want now, all you need is enough tokens
Parkside Picture Poker with Arnold Schwarzenegger
>>110135172lol its trash, just run NoobAI/illustrious locally, 8gb vram max, you can also quantize it to 4gb vram with minimal quality loss with forge.
>>110135231i made the OP a little shitpost-y honestly, cause i didnt really expect this general to actually blow up, but yea we need a new OP like this
>Shutoko like highways>Hardcore simulation physics
>>110135482oh wow, compared to how it was years ago it really has changed. Is it a hassle to set up? Not that but I mind but just wonder.Now for Loras do you just look for some or you have make your own with a style you like?>>110135525Nice I thought it would be like a 5 minute wait for a gen lol
NIGGA WE'RE SO BACKhttps://x.com/customuse/status/2113322603717050811
>>110136086Im glad you did cause this is my new favorite general on this board
>>110136226>add traffic for cutting
>>110136086Well, I think it was about time we got something like this, we could also make a interesting news or something like that in the OP for the current links being postedlike>>110136254And whatever other stuff that pops up
>>110136254okay but I'll see what it's worth when it's a fucking product, enough of this twitter shit>>110136226>>110136295record it in movement bro, it's a car game, we can't see shit
>>110136254Realistic humans in browser is around the corner. Check out this thread between a guy who made a super realistic human, and a guy who ported Metahumans to browser.https://x.com/tfdev__/status/2102486764204277898If clothes AI, maybe we could make some cool ass sim shit in the future. Generate different face JSON presets, slap on metahumans, generate different clothes for them, etc. GTA 7 IN BROWSER BEFORE GTA 6
Can't wait to do this in 6 months with local models lol.
>>110136682what's happening in 6 months anon?
>>110136862We wait just 2 more weeks.
>>110136343
>>110137450Pretty cool anon! What are your plans with this?
>>110137611right now the plan is to be lighting up all those tens of thousands of lights when it cycles from day to night
>>110137756Needs some stakes nigga, have a biker cop or batman or some shii pursuing
>>110135172its getting shilled in every other thread that mentions AI, while saying Local is bad. they might actually be running a campaign lmao
>>110137756Do you plan on adding damage models to the cars?
HI I AM PORTING ZELDA 64 TO THE GBA IN 3D WITH GEMINI AND PYTHON. BUT WITH MORE FEATURES LIKE MEDALLION USE, YOUR BOW IS ALSO MEDALLION BASED NOW. SHADOW ARROWS ARE STEALTH, SPIRIT ARROWS ONE SHOT LIVING THINGS, NO DAMAGE TO DEAD FOR EXAMPLE AND SECRET RELICS AND SKILLS FROM OTHER ZELDA GAMES. I WILL RELEASE THE ROM AS I FIXED IT TO WORK ON REAL HARDWARE SO I CAN GET FEEDBACK SOON
I have a neurological disorder that is preventing me from doing much thinking. Maybe AI with let me escape tactics hell what's a good [spoiler]free[/spoiler] place to start?
>>110135256Why play this over forge? Unless the very limited cardpool is a feature (it is)
planning to do my own llm
>>110136251>Is it a hassle to set up?Only for comfy. But then you can point local agent to fix your shit because it even has an mcp now. Or you can use forge neo for a more conventional UI and simple install.As for loras, chances are you'll already find the artist or style that you need. Either on civit or other websites. But it's pretty easy to train your own. Collecting the dataset for training is the only hard part.
>comfy>look inside>actually hassle
How much do you guys spend on AI a month? What's a reasonable amount to get started with rather than putting too much in and not using enough of it?
>>110139961200$ claude is going to be too much for most people ITT unless they're doing it full time and work on multiple projects at once, I can't fill my weekly quota of it personally even though I'm spamming opus xhigh for 3 projects all day everyday.Codex is absolute trash, don't spend more then 20$ and only do so to get access to the gpt images. The constant reseting may seem appealing; but sol is... not better than sonnet and more expensive to run still, but you get resets so it's a bit of a 5050.Google's argon seems to be just as expensive as astra is, and the performances don't seem to be much better than astra either, but I only tried it on a 20$ plan. Most people can probably make do with one claude 20 for the big smart jobs, and one other (another claude, or a gpt/antigravity) that would be used to do most of the brunt of the work.I'm hoping the next generation of chink models will try to close the gap with opus, I'd fucking love for a cheap glm 5.4 or deepseek 4.2 that'd be in the middle between opus and their current performances, then you'd be able to use those for 80% of your gamedev needs (opus is overkill for them, tbqh).
>>110131903I was thinking of trying to use AI to reimplement a bunch of old amiga point'n'click adventure games and adding them to ScummVM.
@Grok remove their shoes.
