megachud I found in my hard drive edition/gedg/ Wiki: https://igwiki.lyci.de/wiki//gedg/_-_Game_and_Engine_Dev_GeneralIRC: irc.rizon.net #/g/gedgProgress Day: https://rentry.org/gedg-jams/gedg/ Compendium: https://rentry.org/gedg/agdg/: >>>/vg/agdgGraphics Debugger: https://renderdoc.org/Requesting Help-Problem Description: Clearly explain your issue, providing context and relevant background information.-Relevant Code or Content: If applicable, include relevant code, configuration, or content related to your question. Use code tags.previous: >>109101668
>>109127333have you seen the one where the entire terrain is just signed distance fields?
>>109138028nta, but wouldn't it run badly, especially with deformations? SDFs tend to get slow as the CSG tree grows.
>>109138102been a while since I last saw it but it seems to run pretty fasthttps://www.youtube.com/watch?v=il-TXbn5iMA
>>109138102Dreams is pretty much all sdfs and it runs great on a ps4.
>>109138109very impressive
>>109137873fuck off back to your containment thread
find it really funny that third worlder roach people hate trannies just like chud. ironic
>>109138109Yeah but you can't really escape the goopy look can you?
It's too fucking hot to dev
>>109139688sovl
any tips for best LLM for code?
>>109140969already know how to code and know what good code is. you still actually need the skill
>>109140969>>109141063real answer.you need to treat the coding agent like it's a coworker. These LLMs were trained to act like people. if you don't know how to be a leader, set directions, gatekeep quality then you won't be able to use any LLM effectively.all of the subscription models are good. it's a skill issue.
>>109140969GPT-5.5
>>109141260having claude build me a voxel modeler. very very early stages. a lesson I've been repeatedly learning is the tradeoff between building on top of a vertical slice PoC and a fully thought out design so that the agent only has to build it once.when building on top of PoCs, agents often get stuck churning old shit they should have long since forgotten about, instead assuming they're critical features.no relation to the original game btw just dicking around
I think I may have just discovered a GDC Talk worthy point in the VRAM space optimization I just came up with.Just got a 248 MB Terrain down to 20MB in VRAM....Here's my latest visual progress, I like this vid so I'm posting it again.
>>109140969>>109141063Yeah, it’s like managing a junior dev. If you want granularity you gotta know what it’s doing.
>>109143464Tell us about it.
>>109139688simply up the resolution as technology improves
What is your preferred way to render text?Should i use a simple ASCII bitmap atlas or should i use Freetype+Harfbuzz like a good goyim?what about stb_truetype?Is text shaping necessary? It seems that in languages that matter, glyphs map one to one to codepoints, so you can simply use Freetype w/o harfbuzz. (altough freetype doesnt read kerning tables does stb_image read them?)
>>109147086>Is text shaping necessary?no
>>109143464Are you making Worms in 3d?
>>109147086Just a msdf texture
>>109147086I always generate a font atlas with BMFont and use that.>Is text shaping necessary?don't even know what that is. guess not.
>>109137873Beautiful. I don't have words to describe this game. I need get a time to play it.
>>109147086>languages that matterbrave business decision to constrain your market potential in the conceptual stage
>>109147086there are special combining characters, then there are unicode glyph positioning which is what harfbuzz does.glyph positioning/substitution isn't important, because every language except arabic and indian don't use it.most people don't use combining characters, but they might sneak in, for example you might have a letter with a hat (but I believe all languages with hats on their letters will merge the hat into one letter using dead keys / whatever, because most keyboards are designed/based on DOS, so the keyboard is only capable of outputting extended ascii AKA not utf-8, I think).also fun fact, TTF/opentype files can only store 65k letters, which is enough to store the entire Basic Multilingual Plane + maybe a bit more, which is good enough for most languages without emojis. For chrome to handle every character, it needs to chop up the font into multiple files, such as noto font.Also chinese and japanese can't share the same font. If you use the wrong font they will notice (but CJK fonts require an IME which is difficult to handle, also on windows, depending on your version / graphics API / driver, you can't use an IME in fullscreen, you can open the emoji IME using windows key + period or semicolon).
>>109148479>you can't use an IME in fullscreen*it will work, you just won't be able to see the UI, which a native should be able to navigate easily.Also JP has no spaces between words, so if you had word wrapping, that's a small issue to handle (there is a boost locale feature that's able to lookup the JP words so it wraps correctly, but it requires ICU which I remember being very annoying to build, but now that we have vcpkg, it should be easily to get it working, however I wouldn't be surprised if it was like 1gb+ to build, and adds 10-40mb in DLL's).I use unifont and parse the hex font as a fallback, but it is annoying to align TTF fonts with a bitmap fonts. And the hexfont is 8mb to store every single letter, and I would need a 2nd copy for chinese (if I compress, 1mb). And it's not trivial to load every single letter onto the GPU quickly without wasting too much GPU vram (I lazy load into ram so that it's O(1) lookup using a chunked structure since 99% of the unicode range past the BMP is empty, and upload each letter one by one to the GPU).This is a neat atlas packing algo, I use a worse approach:https://osor.io/text
Retard starting from nearly 0 knowledge, working in unity and currently learning to implement the ability to collect things. Demo in 2-3 years, it will be fun I promise. Wish me luck. Good luck everyone.
>>109147086make a shader where every symbol is defined with 3D geometry equations
>>109149063dynamically compile fonts into shader code on the fly so that way you don't have to manually define symbols
>>109145808Terrain voxels are just 1 bit in a bit mask per brick in the brick tree.Structure voxels are the ones which are 1 byte , carrying a material id. The Terrain voxels are painted via a biome palette catalog indexed by biome id which is per instance.>>109147107I'm making a game similar to Dwarf Fortress but with way more to it. And also other games.
>>109137873Has anyone here tried Defold? How would you say it compares to Godot? (for 2D games ofc)
3d enginedevs, what's the best approach for non-realistic physics like character movement?Specifically, suppose I have a character modelled as a capsule or cylinder. I need precise collision detection and resolution between that capsule and a large static world of triangle meshes, plus some dynamic cubes/cylinders.The character should have some vague sense of physicality to it in the sense of accelerating from a stop, sliding down steep slopes, not sliding down mild slopes, etc.The player never bounces, and never rotates except around the up axis (and maybe to follow slight slopes, but that can probably just be visual).Dynamic moving cubes should have a basic effect on the player, including moving platforms. They don't need to affect each other, and the player doesn't need to affect anything in the world.Overall the kind of movement required for N64/PS1 level platforming or a late 90s FPS.Do I want to use a physics library like Bullet/Jolt for the sake of detection/resolution, and just follow their examples for non-physical character style movement? Is it realistic and sensible to DIY it?
>>109149919Use continuous collision detectionPhysics libraries tend to focus on rigid body simulation, some of them have limited CCD raycasting functionality
>>109148914Don't put yourself down, bro. You'll make it if you try.
>>109148914Good Luck to you as well
>>109149919if you already have a custom map format + a way to edit and you already have something on screen, then yeah, you can probably do the physics as well.you are already further ahead of 99% of people in this thread.you can borrow some of the code that you use to cull the view frustrum for your physics, obviously you won't have a frustrum but you probably have some sort of structure for your map. Then you apply your favorite physics algo. (This could be faster but it's unlikely to be an issue if your game is made of large ps2/n64 shaped triangles, you might even be able to brute force every triangle against your character if you only have like 1000 colliding triangles for the whole map)if you have nothing, use unity or any other engine. And just make a prototype in a week.C++ programmers have the hardest programming language, if you can't boot up unity and make a demo in a week, you are just retarded and C++ doesn't make you smart, it just makes you feel smart.>I spent too much time in C++ and I am forcing myself to find things that engine devs can't do, which is a really hard task and it limits the type of games that you can make to a very short list, because unity/whatever is engine dev if you try hard enough.
>>109150049an entire post of anti-advice
>>109149983>continuous collision detectionMy understanding was that this is more of an issue for things like bullets. Is it as much of a concern for my use case (relatively large and slow-moving character)?>>109150049Thanks anon. You've giving me slightly too much credit (no real editor yet) but I do have some objects in a world that get drawn.Using unity/godot would be much more sensible in every way but I have the same curse as probably all the other enginefags... I just don't enjoy using them and need to handcraft the details to get a sense of control and ownership.
>>109150144CCD algorithms give you perfect results as opposed to regular algorithms which give you results that work most of the time
Has anyone had success with AI creating good models? I've tried using Opus 4.8 with the Blender MCP, but it's produced a less than okayish results (pic).I read that meshy.ai might be good. I'll try that next.
>>109150928meshy.ai turned out okay. I'm going to try TRELLIS+Hunyuan3D next.
>>109143464Lords of Uberdark, but maybe will get further than a kickstarter scam and a shitty alpha.
>>109150928They're all mesh nightmares, but for my usecase, it's not a big deal. Problem for me is automating the animations. I'd show you my Codex+ComfyUI MCP+Blender MCP workflow, but it keeps running out of memory alongside my RL testing of my SRPG game engine.
OFP on iOS
>>109151794>spyOScringe
>>109151689I need to do the animation piece soon too. I suspect I'll have problems. TRELLIS does okay with the meshes.You prefer using the Blender MCP compared to any of the specific 3d asset generation models?
>>109151976Haven't had the chance to try more than a few and there are a lot of them. Most were pretty good, but you have to pay to really test them.All my work is 3d to 2d rendered spritesheets. I had more success asking Codex directly to generate assets over anything else so far. If you find something that can generate animations from descriptions, I wanna be the second to know.
>>109151976Oh, I'm dense. I meant that I used TRELLIS2 inside ComfyUI.Missed putting that in here >>109151689Blender MCP makes very basic shapes scattered around, but it's very good at taking one image and building a general shape that you can play with. Boilerplate stuff, really. Not much to work with compared to TRELLIS2.
And I mean very, very rough general shape.
>>109148479>TTF/opentype files can only store 65k lettersOH in 2023 harfbuzz 8.0 added in beyond-64kand also wasm for the shaping VMhttps://www.phoronix.com/news/HarfBuzz-8.0But I can't find any fonts that actually use this.
Cool AI slop guys
>>109152356this may be the only reason they're giving us GTA 6.
>>109152371I don't think the guys at Rockstar are prompting 3D models into existence
Thoughts on Unreal deprecating blueprints?
>>109152376So that's why it's taken so long.
>>109152401good, but unfortunately for the wrong reasons because Epic fell for the LLMeme. So codebases are gonna get even worse because Verse is a shitlang.
>>109152440the Verse language actually looks pretty interesting
I'm a complete newfag to game developement, are there any crash courses I can take, or any software I should use over others? I know it's a very vague question but how do I know what programs are better than others for certain things?
>>109152708learn to use Unity
>>109152708Ask AI.
Anyone have experience making an aiming system in 3D? Specifically the animation side? I basically rigged up a character then used inverse kinematics to control his spine for vertical aim, then I just rotated his whole body for horizontal and those work together to make the gun point at the mouse position, but I’m getting some terrible jitter issues. How do you approach this?
>>109152948that's a fine solution
>>109138265It is you who doesn't belong on this site. You decided to come here and become one of the "cool kids". So now you shut your mouth and blend in instead of asking everyone else to accommodate your mental illness.
anyone got a tldr on how to local llm for devving, I want to check it out
>>109152136why the extra wheels in the tracks that dont touch any of the trackalso why's the track slack on top
Work on Dreadlines (my monster girl dating sim set in an office) is going well! I hired a background artist and finalle found a UI I like. I plan to have a demo out by October for the Nextfest
>>109153553finally* whoops
>>109153514Search on ollama for one tagged “coding” or “tools”, setup the http api, then plug it in to vscode
I had an idea for a 3D level editor, everyone loves minecraft but the problem is you can't even make quake style levels because of the limitations of the cubes. But what if you just had another brush? Like a diagonal brush, like marching cubes? And you had both and you could switch between them instantly?
>>109154321unreal engine had some level making tool like that when I tried it a couple years agothe basic shape was a cube on a 3d grid but you could change the shape and grid size
https://www.youtube.com/watch?v=7e90ZLa5ErA
>>109137873>no mednigger shitalian enemy>no HIVan enemyShit game.
>>109154722So which one of the existing 8 are you?
>>109154321>Like a diagonal brushThat's slopes.>everyone loves minecraftNot as a tool to make actual levels for games.
>>109153964I use intellij's shit
