[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/v/ - Video Games


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


anything above this polycount is lazy unoptimized jeetslop
>>
Fact
>>
>>724628340
>tris
But I see quads...
>>
>>724628761
A quad is just two triangles that share two vertices.
>>
>>724628340
>no fingers
fuck off boomer
>>
>>724628761
only faggot cares about tris and quad. You can have tris/n-gon in your quad model, its fine.
>>
File: 1698219028892163.png (353 KB, 512x512)
353 KB
353 KB PNG
>>724628761
There's actually all kinds of muscles here!
>>
>>724629183
These days you'd have to program your own engine to even end up with n-gons. All of the popular commercial engines triangulize your meshes upon import.
>>
>>724629106
>no fingers
>no eyeballs
>no mouth
>full tits
dangerously based
>>
>>724629290
all meshes are converted to triangles in the GPU anyway, GPUs can only rasterize triangles, and for good reason. A quad/ngon is nothing more than multiple tris where how it's divided into tris isn't explicitly defined. The only reason that 3D modeling memes about quads is that quads are easy to subdivide, and things like generating edge loops and subdividing play much nicer with quad topology because it's very clear how that subdivision will happen.
>>
File: dqviii.png (936 KB, 1737x1336)
936 KB
936 KB PNG
Nah, give artists something to work with.
3,000 tris.
>>
File: kotor1 3.jpg (1 MB, 1920x1080)
1 MB
1 MB JPG
playing kotor1 for the first time, game from 2003 still looks cute
>>
>>724629526
I know. I actually tried to verify the other week whether or not the urban legend about Sega Saturn being able to render quads was true but I found no evidence to support that.
>>
>nanite
Heh
>>
My models have about 1.8k each.
I failed the low poly aesthetic
>>
big ol' dumpy
704 trangles. Could have been much fewer, but I just needed something kinda low poly that would apply to existing 1998 mirror-UV'd texture maps, so I needed versatility for the texture islands. And I couldn't be bothered with manually assigning fancy vertex normals so it's all automatic smooth shading
>>
>>724629717
It seems to be true, but the problem is the Saturn hardware didn’t rasterize arbitrary quads. It only rasterized specifically planar quads. The reason triangles are the primitive of choice for all modern rasterization is that 3 points fix a plane; there is no such thing as a nonplanar triangle. But nonplanar quads can exist; 4 points are not guaranteed to be coplanar. Hardware that can rasterize “quads” well specifically is great for 2D graphics because rect/box shapes are the natural primitive to use for 2D assets. It seems like this is what the Saturn optimized. But that’s not what 3D modelers mean by quads; if you were rasterizing true 3D graphics on the Saturn you’d run into issues when any truly 3D shape was composed out of quads and on the back-end the rasterization process would fundamentally devolve into triangles anyway (because at the end of the day rasterization requires a clear definition of a pixel’s location in 3D space, and with a nonplanar quad, you have to essentially pick a position for a pixel on a nonplanar quad based on an assumption about which tri it should belong to)
>>
>>724628340
Even though you can compensate ultra low poly counts with normal maps, it's not going to work for characters because they need to be deformed (animations). The right answer is to make environmental props in high poly, decimate to low poly and bake normal maps from high poly, and for characters, just keep them high poly. That's if you want high fidelity with good performance, and if you're going for lowpoly style specifically then yeah, do whatever.

Screenshot: the models on the left are 70000+ polygons each, the models on the right are 256 polygons each, with normal maps baked from models on the left.
>>
File: 254.png (277 KB, 610x754)
277 KB
277 KB PNG
>>724628340
>>
File: nhb45y4u8r.png (665 KB, 932x524)
665 KB
665 KB PNG
>>724630076
>>
>>724630210
Ironically the opposite will be true in most cases. GPUs are really, really good at rendering triangles, so much so that you can essentially full-pack your screen with tiny 2x2 pixel quads and experience no meaningful framerate loss from rendering a single quad covering the entire screen. Using normal maps specifically to capture details at a resolution coarser than about the individual pixel is really just wasting VRAM/texture memory. Granted overdraw becomes an issue with high-poly environment assets, but this is better remedied with LOD than with normal mapping.

Meanwhile, when it comes to deforming objects, vertex deformation has to happen on the CPU, and the cost scales as the product of "number of vertices deformed" x "number of bones influencing the deformation". So the more verts you put on a deforming object, the harder the CPU has to work to compute the final vertex coordinates of the deformation for the GPU. CPUs aren't as optimized for this process as GPUs are for rasterizing triangles (they can't be, CPUs are by design highly serialized compute architecture).

Normal maps are great when you want to communicate a very fine texture, such as the rough surface of sand or something. And if you're already USING normal maps for surfaces, then baking details into those maps is essentially free, so you can squeeze a few spare tris out of the process. But it's a weird ass-backward habit of most modern devs to nuke their VRAM usage with 2k normal maps on every object to save 20k tris, when the modern GPU (and by modern I mean basically everything starting from a 970 up) is way more limited by VRAM available than by fragment shader time.
>>
File: DUSK_scarecrow.png (1.24 MB, 1460x714)
1.24 MB
1.24 MB PNG
>>724630289
One thing that I absolutely despise is when fake retro developers make low poly models and then apply either sharp corner shading as if gouraud didn't exist in 1995, or automatic shading, mening that every angle below 40° is sharp because the modeler is literally an amateur who doesn't know what vertex normals are... and all of that is then packaged into le epic retro graphics PS1 anal analog horror indie gem! Displaying technical limitations that never exist in the form depicted, and going beyond the limitations of the time it pretends to emulate.
Even DUSK is guilty of this. Triangular barrels with sharp corners on guns even though skinned mesh characters have smooth shading, as if the renderer is simultaneously capable and incapable of vertex-shading corners.
>>
>>724630664
>vertex deformation has to happen on the CPU
Wait... is that the reason Elite Dangerous Odyssey takes so much CPU and why framerate improves as you kill off NPCs in a base? Those skinned meshes are no longer receiving deformation calculations and IK solves on the CPU? I wouldn't be surprised if those meshes don't have any LODs that would lessen the vertex deformation load by limiting weights and that's the reason for the framerate differences
>>
File: DrawCallPerf.png (1.33 MB, 1881x1054)
1.33 MB
1.33 MB PNG
>>724630874
Possibly, I'm not an Elite Dangerous expert. It could also be related to badly optimized AI behaviors (the more enemies you kill, the less work there is ticking the AI behaviors of various agents, if you're doing a whole lot of agent ticking especially when those agents aren't on-screen you can quickly explode your CPU time).

I can tell you that this effect is why you can get excellent FPS with tessellation turned on (turning every flat quad into a dense mesh of totally redundant tris) but the un-retopo'd dragon from Forspoken tanks the FPS to single digits. Because the dragon is trying to compute smooth gradient bone deformation for those millions of verts, rather than just passing them off to the fragment shader without asking questions.

I've posted this pic before from when I was doing some profiling of my own game, where I demonstrate that I go from rendering 6 million tris to slightly under 4 million tris and LOSE 10FPS, and the reason for that is because the draw call count quadruples from 400 to 1600 (draw calls are CPU bound). And this is on a GTX1070 for fuck's sake. And that's with only 170MB (roughly) of VRAM utilized; if you start hitting the swap point on your VRAM limit, god help you.

The point is, GPUs are architecture designed to rasterize millions of triangles quickly. They are very, very good at doing it. Most major framerate struggles come from situations where you aren't allowing them to do that; either because you are making them sit around and wait for the CPU to send instructions, or because you are forcing one rasterizer to wait on a previous rasterizer (i.e. lots of overdraw).
>>
>>724629590
Too much.
>>
>>724630664
> GPUs are really, really good at rendering triangles
Yeah, great, lets keep rocks 70000 polygons each and put 10000000 rocks in our open world game.
> Using normal maps specifically to capture details at a resolution coarser than about the individual pixel is really just wasting VRAM/texture memory.
On my screenshot normal maps are 512x512, so they are not coarser than individual pixel.
> Meanwhile, when it comes to deforming objects, vertex deformation has to happen on the CPU
This is only the case in Blender and it's almost considered a bug in Blender that they still can't animate on GPU.
> But it's a weird ass-backward habit of most modern devs to nuke their VRAM usage with 2k normal maps on every object to save 20k tris
They can tank VRAM on older GPUs by being overly wasteful with normal map resolutions, but if they just try to keep geometry, it's probably not gonna work at all, on any GPU whatsoever, because when you save 20k tris on environmental prop, you're not saving it once, it's about doing the same to every environmental prop in the scene, which sums up to dozens or hundreds of millions of polygons.
>>
>>724632509
>millions of polygons.
See
>>724631227
Millions of polygons really isn't a problem.

>This is only the case in Blender
No, vertex deformation is always a CPU side process. Vertex animation CAN be done on the GPU in the vertex shader, but I don't know of any engines that compute bone deformation in the vertex shader. Skeletal animation is computed CPU side.

>10000000 rocks in our open world game
You can LOD those rocks to like 6 polys once they're far enough away from the camera and it stops mattering. The real reason to avoid triangle density is to avoid overdraw (you don't want one pixel's value to rely on the rasterization of 500 triangles, because that means the GPU has to do 500 serial rasterizations to shade that pixel, it can't do it in parallel), LOD is your saving grace here, not normal mapping.
>>
>>724630720
I like automatic shading by angle in lowpoly context, it gives this unique style of sharp edges gradually blurring out into the surface. I've never seen this used in games. It's kinda hard to work with so I'm not sure I'm ever gonna make a game in this style, but it's definitely something I wanted to try for a while.
>>
File: mms.png (38 KB, 124x120)
38 KB
38 KB PNG
can any unreal engine fag tell me why are masked shadows so expensive in this engine? making a simple tree with 20 overlapping foliage masked sheets and enabling dynamic shadow cast on it is goddamn expensive but having every leave modelled and casting opaque shadows is very cheap in comparison even if it has 10 times the polycount
>>
File: Capture.jpg (411 KB, 2268x1264)
411 KB
411 KB JPG
No. The simple necklace should have so many triangles on the chain alone that the model viewer doesn't show the count.
>>
>>724632712
So I did a bit of googling and it turns out that actually, a lot of engines can do a kind of "hardware skinning" in the vertex shader, just with the caveat that your maximum influence-per-vertex is four bones, and more than that and the mesh must be skinned CPU side, because that's all the hardware is optimized to handle. Which makes sense, past a certain point the effort involved in computing redundant bone transforms for each vertex outweighs the savings of parallelizing it. But that's kind of neat. It would still be true that many modern rigging practices would make this not work terribly well, but it does mean that certain kinds of rigging (e.g. rigid body rigging) can be offloaded to the vertex shader and not impact CPU time.
>>
>>724632712
> No, vertex deformation is always a CPU side process. Vertex animation CAN be done on the GPU in the vertex shader, but I don't know of any engines that compute bone deformation in the vertex shader. Skeletal animation is computed CPU side.
Ah right, I confused that with GPU skinning. There are many 3rd party tools for GPU vertex animations for Unity, but they seem they don't support blending well.
> You can LOD those rocks to like 6 polys once they're far enough away from the camera and it stops mattering.
Okay, why not use LOD for normal maps? This whole argument is more like: "if you don't have too much objects in nearby range at any given moment, you might skip using normal maps and just keep high poly counts for environment assets".
>>
>>724634015
Overdraw is the most likely culprit. If you turn on overdraw visualization on your masked-mesh tree you'll probably see a glowing-hot cluster of bright white. 20 overlapping foliage sheets means each pixel must wait for 20 discrete GPU cycles to complete; you cannot rasterize any given tri properly until the tri beneath it is rasterized when you have transparency/alpha masking. Then consider that shadow-casting has to do this same process but AGAIN for each light source in order to compute the dynamic shadows; each light "ray" simulation must compute 20 layers of stencil to get the final shape.

Meanwhile if every leaf is a properly cut-out polygon, a lot of that overdraw is gone, because you don't have to rasterize "invisible space" 20 times to calculate the pixel value. Sometimes you still do, when there are actually 20 leaves overlapping in a space, but that happens a lot less often when you don't have a bunch of wasted transparency.
>>
>>724632898
No I don't mean smoothing groups. Those are a good and often necessary practice. For example a normal map can't properly shade a sharp corner to look like it has a bit of a bevel, because the baking requires a margin. What I mean is a 3D modeler who does not know what sharp angles and smoothing groups are, and just leaves the vertex normal and sharp angle definition auto-angle as the default, meaning that all the sharp angles are just what the program detemined should be sharp, whether you meant it or not.
>>
File: p10.png (1019 KB, 800x1200)
1019 KB
1019 KB PNG
i used 4.9k tris on this one
>>
Heckin based fren
>>
File: brave_3PBQDk2zxy.png (824 KB, 1053x818)
824 KB
824 KB PNG
We need to put character artists back in the texture painting mines.
>>
>>724634552
>Okay, why not use LOD for normal maps
You can do that too but the issue is not the compute cost of the textures (texture lookups are basically free of charge), it's the memory required to store them all. You'd get horrendous lag spikes if you unloaded a texture from VRAM every time the object got too far from the camera and reloaded it when it got closer; if an object is loaded, you've got the texture data for its highest fidelity rendering in VRAM, even if you aren't using it. At least in most cases. There are probably texture-streaming solutions designed to work around this but that kind of tech isn't really common-place because high-speed data streaming to GPU requires enormous SSD bandwidth.

Granted the same is true of vertex data, of course. You're holding the object in memory meaning you're holding the vertex data for every LOD instance of the object in memory. The difference is that a single vertex uses about the same amount of data as a single texture pixel (a 3 vector); so a single 1024x1024 texture in VRAM contains about as much information as an object with 1 million verts

not really, because textures are compressible, and also because vertex information has a lot of redundancy due to how triangles are specified, but it's still generally true that storing vertex information is way cheaper than storing texture information relative to the fidelity per pixel on screen
>>
File: low poly2.png (203 KB, 815x821)
203 KB
203 KB PNG
low poly oozes soul and last forever
photorealism is slop and ages like milk
>>
If you need more than 50 tris to convey your character then youve already fucked up
>>
>>724634165
jeez, whats this from?
>>
>>724634869
The thing with rocks, for example, is that keeping this detail in geometry requires extreme amount of geometry. I can't really go from 70k to lets say 4k, going from 70k to 256 + normal map looks better than decimating from 70k to maybe even 16k for example. I don't know how it translates to big scenes exactly, but given that the scene with lowpoly style houses in >>724631227 is 4kk polys, a nature with sculpted rocks and trees bark, and stuff might hit hundreds of millions easily I guess.
>>
>>724635061
https://www.nexusmods.com/skyrimspecialedition/mods/126514?tab=files

The author refuses the optimize them because apparently it would break his wifes vision.
>>
>>724634865
Too bad the few that still do it lack their own style and end up looking like WoW or fucking LoL.
>>
>>724635334
Eventually vertex data alone will blow out your VRAM and you will be required to stream data (see: Nanite in Unreal Engine) but you're talking about close to BILLIONS of vertices to hit this point. Take any random Quixel Megascan asset and you're looking at like, 500k plus tris for something small like a rock or a traffic cone or whatever, and when you have tens of thousands of those props all on screen at once, yes you can hit hundreds of millions or billions of verts.

But the real lesson here is that wasting detail on minor environmental assets is stupid. Yeah, you don't need 70k unique tris for every random ass rock and tree in your game. You also probably don't need a unique normal map for every random ass rock and tree in your game. Smart devs are making like, 5 rock assets and then using basic transforms like scale and rotation to make them appear as a bunch of unique rocks. This is why I always roll my eyes when people meme about shit like
>shove bookshelf into the ground and call it a table
in Skyrim or whatever; that's actually really good technical artistry. Reusing models reduces VRAM usage and can make draw call batching possible, it's at the heart of most modern foliage rendering tools, NOBODY is going to notice if your 350 rocks on a mountainside are actually the same 5 rocks with randomized scale and rotation.
>>
File: 20251031_085034.jpg (1.44 MB, 2208x2074)
1.44 MB
1.44 MB JPG
>>724634659
Nice.
I hit, rounding up, 2.5k on mine.
For the sake of coom, i want to make DoA3 models, ill have to up to 10 to 20K eventually.
Textures do a lot of heavy lifting, but to get them jiggle physics, squash and stretch, its going to take more.
>>
>>724635912
I mean, its possible to get a good low poly bounce, but id like a little more.
>>
>>724630664
> vertex deformation has to happen on the CPU
erm don't you upload the skinning matrices for each bone to the gpu?
The vertex deformation itself should be gpu accelerated, only the rotation of the bones themselves would be cpu-side, I thought
>>
>>724635624
>Too bad the few that still do it lack their own style and end up looking like WoW or fucking LoL
99% creative work in the industry looks like this:
>here's a concept art
> nice nice, but can you look more like [the current popular thing they want to copy]
>okay. Here's an update
> nice nice, but can you look more like [the current popular thing they want to copy]
>o-okay
Until you send them an EXACT COPY of WoW or LoL. Woke managers are also awful because they force you to make your art ugly. Now that's AI is blooming, they fire any original or creative artists and either get proompters for concept art (generate a copy of wow concept art) or Indians for 3d models.
The future has never looked so grim, derivative and uninspired.
>>
>>724636056
Yeah I was slightly mistaken, this is in fact done as long as there are few enough bones influencing any given vertex. You can index bones and then pass the bone transforms along to the vertex shader in index order and compute them there, but the hardware can only accelerate it when there are a sufficiently small number of bones per vertex. If you have 5 bones influencing the position of a vertex, you have to do the vertex compute for the entire model on the CPU, because GPU architecture is only designed to handle the cache information for 4 bones per vert. Why exactly that is, I have no idea, presumably it's some sort of bandwidth thing where you have exponential scaling of how many weighted matrices you have to consider per vertex, but the hardware architecture sort of is what it is at this point and 4 bones per vert is the magic number.
>>
>>724628340
I've been thinking, are triangles like the ones in that model, on the shoulders and waist and shit, a good solution to the wrapping edge loops problem? I never could figure that one out.
>>
>>724636540
>>724636056
Actually damn, the more I look into this the more wrong I see I was, it seems modern pipelines can handle more than 4 if you ask them to. UE for instance can hardware skin 8 bones per vertex. The thing is, I don't really know how this limit is engaged; if you go above 4 verts per vertex, presumably it has to increase the buffer size of the entire rig? Meaning it increases the compute time on EVERY vertex, no matter how many bones influence it, because it still has to do a 5x or 6x or whatever matrix operation on every vert due to the new index size? I mean it can't use an arbitrary matrix size per vert, that hits the point where the math would be so complicated it would just be cheaper to do it on the CPU to start with.

It's interesting and obviously I don't know enough about how skinning is handled in modern pipelines.
>>
>>724636540
Modern shaders are very flexible, you could certainly code gpu skinning with 8 bones if you wanted, but performance would be worse and you would need to bloat the vertex structure with more matrix weights etc.
In practice, in any normal skeletal mesh, there's almost never a need for verts to be influenced by more than 4 bones anyway, though.
>>
>>724635848
When I'm discussing this topic, I'm ofc always assuming assets are reused. 5 is bullshit, you need at least a dozen variations of each type of prop per biome for it to be good, and you need those different for different biomes for them to feel different enough.
>>
>>724636973
>there's almost never a need for verts to be influenced by more than 4 bones anyway, though.
I think it kind of depends on how you're rigging. If you're modeling a human, PROBABLY not unless you're using stupidly detailed bone structure. But for something that moves like a snake or an octopus, where you're approximating a gradient of curvature using a series of rigid bones, I can see why you'd maybe want more than 4 to influence a vertex, especially to avoid self-intersections when the curvature becomes very tight.
>>
>>724637229
>you need at least a dozen variations of each type of prop per biome for it to be good
Depends on the prop. For rocks, no way. For trees, maybe so, though I think you'd be surprised how far you could get with half a dozen. And when you're talking about "different biomes" consider you probably aren't rendering all your desert assets and forest assets at the same time, so you're less concerned with it.
>>
>>724636652
why would you care about edge loops if its lowpoly? you want to subdivide it evenly lol?
>>
>>724637705
Edge loops at any poly count ensure nice deformation as well as make it easier to edit and adjust
>>
>>724637705
I'm not talking strictly about low poly, just modeling in general.
>>
>>724636241
artists that can paint in WoW style are the talented rarest beasts out there, retard
go shit out some generic rifle or car again, maybe you'd feel better again (not)
>>
>>724637798
everything deforms in triangles, you can't deform better than that
>easier to edit
if its lowpoly the trade off is that there isn't much to adjust
>>
>>724628340
goddamn that's bad topology
good luck deforming that with a rig
>>
File: 20251030154218_1.jpg (411 KB, 1920x1080)
411 KB
411 KB JPG
>>724637361
You must have really low standards for rocks.
>>
File: 20251011192945_1.jpg (235 KB, 1920x1080)
235 KB
235 KB JPG
>>724639973
>>
File: PIleOfRocks.png (167 KB, 927x569)
167 KB
167 KB PNG
>>724639973
I guess for really gigantic pieces of geometry like that where it's actually informing the shape of the terrain, I was thinking more of basic decorative rocks. Pic related for instance.
>>
File: 3dlowpoly.png (384 KB, 734x840)
384 KB
384 KB PNG
>>724634631
well i dont know how to fucking model bro
what the fuck do you want
>>
>>724640649
How the hell do you make so good looking pile of rocks? No matter what I do they look terrible.
>>
Ah yes, low poly. The Redditcore "le artistic""choice"""

Sucks to suck at real modelling lol
>>
>>724642131
Imagine being so dumb you intentionally shrink your possible customer base by making graphics that can run only on high-end computers.
>>
>>724641514
That's literally one single rock mesh, that I just duplicated, and randomly moved, rotated and non-uniformly scaled. That's the thing about rocks and assets like rocks; when the object in question doesn't have any inherent sense of orientation or proportionality, you can get a lot of "bang for your buck" instance-wise by just fucking around with the transform. The underside of a rock might as well be a new rock, nobody is memorizing the shape of rocks that well and nobody notices if a rock is stretched out to be 4x as wide and 1/3rd as tall. especially if you use some sort of global texture orientation, which will hide the inconsistency in the rock texture and will also allow rocks intersecting other rocks to appear as a single rock (since there will be texture consistency throughout).
>>
>>724630076
BRAAAAPPPPPPPPP
>>
>>724642494
I have never made a single good-looking rock in my entire life. They all just dont look nature-y eough to start applying the tricks.
>>
File: 1735274092723510.jpg (615 KB, 1556x950)
615 KB
615 KB JPG
It's wild how much textures can carry low poly.
>>
>>724631245
>15 of the 200 polygons just to give the soldier a bulge in his pants
based Kojima
>>
>>724642131
>real modelling
comeback when you start doing one, sculptlet
>>
>>724642641
I made that rock in a few seconds to prove the point.

Default blender cube
subdivide twice into 4x4x4 vert cube
Subsurf modifier
drag vertices in random directions
Apply subsurf modifier to mesh
Proportional editing of new verts with noise gradient and just randomly yank until it has very little symmetry.
A couple steps of Smooth Vertices to smooth the rock out
Smooth shading on

You'd need a slightly different workflow for large jagged rocks, those look worn like pebbles near a stream.
>>
File: 1726417322324532.png (1.35 MB, 1623x898)
1.35 MB
1.35 MB PNG
>>
>>724642663
her face probably has as much tris as the whole body
>>
>>724642436
>anything higher poly than ps2 requires a high end computer
lol
>>
>>724642826
I was just thinking of her when I looked at OP. She has the perfect female model polycount.
>>
>>724642436
the 90s want their shitboxes back lil bro
>>
>>724642803
>until it has very little symmetry
So thats what I was missing, couldnt really put a finger on it. Thanks a lot, anon.
>>
>>724638323
>everything deforms in triangles, you can't deform better than that
Not true, proper quad deformation in Blender can look a lot nicer than models built with careless triangles
>if its lowpoly the trade off is that there isn't much to adjust
If you ever worked in low poly you'd know there's far more to adjust because you're working with so little you need to constantly reevaluate how you're doing things, like how a pixel sprite can take a longer time than a full on drawing since every pixel matters
>>
File: 20251028221951_1.jpg (250 KB, 1920x1080)
250 KB
250 KB JPG
>>724640649
That's just blobs, if you want to make them look good, you will need to paint all the detail from scratch on your own. If you sculpted some proper geometry there, this geometry would help you with providing you automatic pointiness maps / edge detect maps which can be used to do procedural highlights and edge wear, but in your case all this information has to be handpainted manually.
>>
>>724644486
NTA, whats "proper geometry"
>>
>>724628761
Quads are automatically converted into tris by the GPU. Quads are easier to work with in modeling software so most decent 3D meshes use a few tris as possible and are almost entirely quads.
>>
>>724644828
Look at the stones on the screenshot and realize how certain darker shades represent cavities, how brighter shades represent highlights that suggest sharp edges, etc.
>>
>>724645335
Thats a texture. Geometry refers to the model.
>>
File: 1753105813555619.jpg (141 KB, 984x1200)
141 KB
141 KB JPG
>>
>>724645462
I'd let her sink into my face, if you know what I mean.
>>
>>724645462
thats like 50 zelda ocarina of time's
>>
File: Screenshot (3).png (2.48 MB, 1920x1080)
2.48 MB
2.48 MB PNG
Agreed, but anything other than a texture map such as a normal map is lazy unoptimized
>>
>>724645462
You could play DOOM on that ass
>>
>>724645381
Yeah, with better geometry more of a textures (not only normal maps) are to a larger extent derivable from geometry. This is just a hypothesis but I believe a lot of those stones even in very painterly styles rely heavily on baking maps from geometry in Substance Painter (especially masks that do edge detect) and then painting some extra detail on top of them. In other words, I think those textures might be like 40% handpainted and 60% procedural, and do them procedurally you need this geometry in the first place.
>>
File: soap.png (711 KB, 641x578)
711 KB
711 KB PNG
I don't remember the specifics but the polycount for cod4 was really low per characters. Like sub 8k
>>
File: light_scattering.png (2.31 MB, 1899x528)
2.31 MB
2.31 MB PNG
>>724646510
And get rid of PBR too.
>>
>>724628340
This, Playstation 1 is the greatest time period in the history of the universe and time should have frozen so it could last forever.
>>
File: 1757986927723370.png (294 KB, 454x667)
294 KB
294 KB PNG
>>724646806
>8K
>very low
KEK!
I remember when 1k was "high poly'.
PS2 MGS2-3 had like 3-5k for Snake alone. Half of that for the guards.

Even less on PSP.
>>
>>724647060
PS2 is the peak, even though MGS1 on PS2 outshines many games even on the PS2.
>>
File: ps4.png (135 KB, 673x602)
135 KB
135 KB PNG
>>724647090
Yep, they got too cocky with ps3 and we got slop like MGS4.
>>
>>724647090
>PSP
no one gives a shit about your 400*200 res device
point being a good fucking looking game with great performance that ran in the ps360 era hardware at 60fps and at 300+ @ 1080p on PC is way more impressive on the tech side that your legacy trash from the ps2 era
point being a random twitter tranny can emulate psp era models just fine while not even an indie team can make good looking models like mw1 and 2
>>
File: 4626.jpg (107 KB, 800x518)
107 KB
107 KB JPG
>>724628340
I prefer PS2
>>
File: women.png (1.43 MB, 752x804)
1.43 MB
1.43 MB PNG
>>724647090
I think polycount is not the problem. It's what you use those polygons for. Most fighting games since the PS2 era had high poly models, and yet they aged really well. A mesh designed with subdivision surfaces can theoretically have infinite polygons.
>>
File: 1761584532136218.gif (252 KB, 1200x1100)
252 KB
252 KB GIF
>>
>>724644486
True, and I did it in 5 minutes not because I was creating rock assets for my game, but to demonstrate the principle that rotating and scaling a rock makes it functionally a new rock. That was the point and it holds true even in contexts where you texture crevices and weathered edges and such.
>>
i dont want to learn or do the work, so when will AI be able to generate a fully rigged, authentic low poly character mesh, with great hand drawn (ai generated) textures?
>>
>>724648390
It can do that already, but not very well and you'll still want at least basic competency to clean it up and fix any errors it made, same as AI images.
>>
>>724629598
PS2 era 3D models have massive amounts of SOVL.
I would play SWG nonstop if it had kotor tier character models and animations.
>>
>>724628340
This can't be properly animated, retard. Horrible mesh
>>
>>724628340
Topology is still horrible and unsuitable for animation.
>>
>>724648390
2 more weeks sir
>>
>>724648726
To be fair, you can animate lowpoly with horrible topologies by animating them semi-stopframe style. In Blender terms, after you did the usual animation, you switch keyframe Blending from Linear or whatever else to Constant. Kinda "baking" vertex locations at certain keyframes and then just changing between them immediately. This way you can avoid certain points where there are deformation artifact and only cherrypick frames that look good. I really like how this style looks but it's not very universal.
>>
why is texturing a lost art?
>>
>>724642697
>200 polygons
2253 polygons
>>
File: Screenshot (5).png (2.15 MB, 1920x1080)
2.15 MB
2.15 MB PNG
>>724649116
Who knows.
>>
>>724628340
But how i can make porn with it
>>
>>724649116
Most modern games use minimally altered textures from the 3D scanned IRL objects they're using for game objects.
Indies have a much lower amount of time, likely less access to professional learning resources and feedback.
Its a skillset that isn't needed by those who can afford to make use of it, and so has gone by the wayside.
>>
>>724649352
sovl
>>
File: 1735236094569818.jpg (125 KB, 573x979)
125 KB
125 KB JPG
>>724647864
How do we solve this clipping problem on low poly ?
>>
>>724645462
Fake
>>
>>724649685
https://youtu.be/q7GjGO9nnWg?t=165
Early fighters got you covered.
>>
>>724649116
old devs had passion for their work.

new devs are soulless corpo drones.
>>
>>724648390
You don't, you simply need to kill yourself.
>>
>>724649685
There was some technique I was exposed to recently where you can animate the texture of a characters clothing to effectively hide it or swap outfits. Kind of like some of the tech for swapping low poly facial expressions.
I bet the same thing could be used to alter a clothing texture so it plays nice with certain actions, given modern tech has less limitations and could support the texture sizes needed to support something like that.
>>724649453
that is a good question. I think high quality animation, jiggle physics, and sound should be able to improve the ero levels enough to offset model quality.
>>
>>724628340
3k polys max
focus should be ass and tiddies, maybe hair
>>
>>724628340
Upgrade your computer
>>
File: 5gvy3f3bhu4.png (375 KB, 352x708)
375 KB
375 KB PNG
>>724649685
by not caring
>>
>>724650254
Real man animate the physics manually
>>
File: 1746543166368392.webm (1.42 MB, 1920x884)
1.42 MB
1.42 MB WEBM
>>724628340
how many polys in GTA 6 Lucia's ass?
>>
>>724650054
not yet i need to vibecode a full game using AI tools from the ground up first
>>
File: 1741054220724942.jpg (141 KB, 512x512)
141 KB
141 KB JPG
>ctrl f normal map
>multiple results
/v/ is less retarded than usual today
>>
>>724650535
you just avoiding the realism sloppa you secretly love for some reason
just go full retard, we don't have technical limitation of 1998 anymore
>>
File: Screenshot (8).png (1.16 MB, 1920x1080)
1.16 MB
1.16 MB PNG
>>724650535
Name one game where normal mapping actually added soul.
>>
>>724649685
I just didn't bother rigging the dress separately from the legs, that could be easily be fixed with more bones
>>
>>724650445
The jiggle motion feels very unnatural and amateurish, I've seen better in indie porn animators which should be surprising for Rockstar but it's not given all the news we've heard about the takeover that happened
>>
>>724650439
I mean, sure, but I'd rather spend that time elsewhere, like getting my character pupils to dynamically look around (and have said pupils dilate if she sees your characters dick)
>>
File: 20250718035838_1.jpg (780 KB, 1920x1080)
780 KB
780 KB JPG
>>724651198
Demon Turf. Adding normal maps into cartoon shit and making it work somehow is chefs kiss. Overall that game is crazy salad of styles and techniques, it's all sloppy and often rough. In a sovl way.
>>
>>724651792
I don't see any tangible difference contributed by normal maps.
>>
File: 20250717191853_1.jpg (335 KB, 1920x1080)
335 KB
335 KB JPG
>>724651903
That whole volume between tiles on the floor is solely by normal maps, it's actually flat. Yeah, that's a very brutal example. They are mostly used in more subtle ways.
>>
File: Screenshot (6).png (1.21 MB, 1920x1080)
1.21 MB
1.21 MB PNG
>>724652952
And yet, i don't think it's contributing anything to the game, you can probably replace that thing with a better texture and come out better.
There's not a single normal map in picrel.
>>
>>724653197
Still looks flat, which is reasonable since it came in 1998 and the Naomi can use bump mapping, at least on environments.
>>
>>724651440
but did you model the pantsu?
>>
>>724653197
normal maps is better for scenes with lighting changes, they dont do much in scenes with static illumination
>>
File: lowP.webm (2.65 MB, 640x1150)
2.65 MB
2.65 MB WEBM
>>
>>724628340
It's not slop just because it's to powerful to run on your steam deck.
>>
>>724651792
I'm actually not sure, it might be parallax. This is the best I could get in terms of raw volume via bump map (in 15 minutes anyway).
>>
>>724629196
Mt. Carlos!!
>>
Can you even get height maps or paralax on hand drawn textures?
>>
>trying to make Blender feel more like Maya
>>
>>724654387
They do a lot, it's just that you can bake them into textures when there are no dynamic lights.
>>
>>724654785
yes, its somewhat tedious, but it can be done
>>
>>724654785
Sure, for starters you can just draw height map by hand as well, or you can use texture, convert it to grayscale and use that as height map. This might be what's going on in >>724651792. The black parts of the original texture are mixed with some extra noise and used for displacement via parallax.
>>
>>724654785
if people can hand draw normal maps they can hand draw height maps
>>
https://www.youtube.com/watch?v=1uFPmmZFyvQ

I'm curious how the lighting and models work in luigi mansion 3. Is it all realistic PBR textures with height/normal maps that has been stylized?
>>
>>724654594
That picrel proves my point. Normal/Bump/Height map aliasing is not worth the effort.
>>
>>724650445
Fuarkk
>>
>>724655201
That smudgy texture aliasing. Yep it's PBR.
>>
>>724654816
You mean bake the lighting into the color texture. Normal map is useless without lighting and you don't "bake normal map into textures", you bake highpoly details into the lowpoly's normal map.
>>
>>724656350
> You mean bake the lighting into the color texture
Yes, yes, I obviously meant that, don't expect people to word everything in the most pedantic way possible like it's some sort of defining dictionary.
>>
File: 006_ac2_051.jpg (105 KB, 718x800)
105 KB
105 KB JPG
>>724643102
That's normal
>>
>>724650535
>seam down the middle
amateur hour
>>
File: 1708349781487120.jpg (74 KB, 720x937)
74 KB
74 KB JPG
>high poly model with separately modeled high poly clothing
nutin personnel toasters
>>
File: 1753229724107292.png (1.75 MB, 2229x1208)
1.75 MB
1.75 MB PNG
for comfy maxxing, put your low poly characters in other vidya
>>
>>724628340
>anything above this polycount is lazy unoptimized jeetslop
glad to see other people agree that expedition 33 is soulless dogshit
>>
>>724649116
a long time ago, render pipelines were fixed, programs could not define custom shaders, only manipulate a few gigantic shaders provided by the hardware. during this time, it was more efficient to buffer in one high-detail albedo than lots of different maps. nowadays, shaders are dynamically compiled per-program. whilst it is still just as expensive to buffer textures through, the advent of dynamic shader compilation was coincidentally the advent of it being such a minimal bottleneck that the buffer time didn't matter anymore so long as you kept it to a minimum per-frame. no need to put a lot of detail into albedos that are baked when computationally shaders can achieve the same thing mathematically. still, aesthetically there is a huge loss of not having hand-painted albedos with art brain detail as opposed to math brain detail.
>>
>>724659551
Look at the hairs in this: >>724650445
Any game that has visible noise is garbage.
>>
>>724660229
>art brain detail vs math brain detail
/thread
>>
File: 1613806556242.jpg (110 KB, 611x674)
110 KB
110 KB JPG
>>724660325
>third option: vs coom brain
higher poly means softer more jiggly body and skindentations
>>
File: TRVTHNVKE.png (2.24 MB, 1050x1177)
2.24 MB
2.24 MB PNG
>>724629590
BS

textures would do a better job than these small 3D bits

2D>3D unless you have extrmeely advanced lightning system and high polycount/shadowing precision

Games these days look uglier than ps3 games for that reason. they used real life textures which leveraged very fine details you couldnt get rasterized/raytraced without advanced shit we still dont have to this day. for wider/more macro details maybe, but not small bits.

games due to this reason look pretty flat and monochrome these days. nothing beats photographs of walls with the dry painting sticking out
>>
>>724660730
Agreed, but PS3 was the beginning of uglification.
>>
>>724634865
LoL style is absolute trash and i dont understand why anyone would copy that yet a lot do it. capitalism and suits whatever
>>
/v/ is retarded when it comes to 3D and this entire thread just proved it.
>3D artist since 3 years in both organic + hardsurface modeling and shipped a game to steam. both lowpoly and PBR
>>
>>724662210
You needed this thread to figure that out?
>>
>>724662210
Let's see it.
>>
>>724628340
how and where do i model like this?
>>
File: Untitled-1.png (701 KB, 2000x1023)
701 KB
701 KB PNG
>>724662619
>>
File: Untitled-2.png (378 KB, 799x594)
378 KB
378 KB PNG
>>724663236
>>
>>724660325
pissfilter is what makes mgs3
>>
>>724630210
Cool. Hats off to optimizers.
>>
File: blender_E86kEVLOFQ.png (720 KB, 1017x686)
720 KB
720 KB PNG
Low poly is nice but obsessing over polycounts in current year is pretty retarded
>>
>>724664569
Correct. You figured out /v/ is retarded. It's not 1998 anymore where a 6000 polygon model can crash the engine. Polygon count literally doesn't matter with todays hardware and engines.
>>
>>724634746
I heard the main Fo3 modeler was pretty inexperienced when he first got in and mostly used to sculpting, explaining why everything looked kind of jank for the year it came out.
Seeing how much he improved in Fo4 was impressive.
>>
>>724664569
>>724664735
>Not going for optimization so you can put together giant scenes
To take full advantage of modern hardware and engines, you need to optimize. God knows how many UE and Unity projects shit themselves on modern consoles because the programmer doesn't know what the fuck he's doing.
The world doesn't need more YandreDevs.
>>
>>724666015
>consoles
you're the reason we went from crysis 1 on PC only downwards to technological de-evolution from that point on
>>
>>724664569
>>724664735
Haha ikr, what a bunch of retards optimizing shit.
Anyways, im still waiting on my 200gb patch to finish downloading so i can play my AAA slop./s you filthy pajeet.
>>
File: 034.jpg (582 KB, 1920x1090)
582 KB
582 KB JPG
>>724651198
Breath of the Wild and Tears of the Kingdom are heavily normal mapped. All the designs you see on structures are normal mapped. And of course on the landscape as well. Every stone in the game is normal mapped.
>>
>>724666578
>update
that´s on you for playing live service shit lmao
>>
>>724649116
see >>724648390
and apply to current year
>>
>>724666771
Not even that retard, games are 200gb plus now a days because of shit optimization.
>>
I just made a nice render, but I can't show anybody. Damn shame.
I'm improving at procedural textures. But I'm lagging in the actual UV and texture department. I pray for some kind of technical breakthrough that makes procedural textures faster, so I'll never have to bake anything.
>>
>>724666578
polycount isnt what's causing that filesize, that anon is right you're a fucking retard
>>
>>724664569
you're a faggot, it's about art and aesthetic which of course a fatfag would know nothing about
>>
>>724668724
show me your so called "art", dear nodev
>>
File: SB_LOD_0.jpg (656 KB, 1874x1786)
656 KB
656 KB JPG
>>724628340
>anything above this polycount is lazy unoptimized jeetslop
Then why does this game run like butter with maxxed out settings on my toaster?
>>
File: GXDEvxYakAA3mC6.jpg.jpg (26 KB, 335x335)
26 KB
26 KB JPG
>>
Running a game with a high poly count is different than modeling or building a game with a high poly count.
>>
>>724662210
/v/'s entire knowledge of 3d modelling comes from a game of telephone
literal cargo cult nigger behaviour
>>
>>724673140
Hey, you faggots chime in with info or fuck off an leave. All i see from your post is parroting like that cult nigger behavior your projecting on everyone else. Prove you know something, else youre just as much a poser as you claim others faggot.
Chiming in like this isnt going to make you one of the cool devs shit for brains. Its an anonomous board, no one knows or gives a fuck about you
>>
>>724672446
its just autists who cant move on or deal with change and want everything to look like a ps2 game, they don't actually care about optimizations, let alone understand how optimization works
>>
>>724673338
Do you have something useful to say?
>>
>>724673140
>literal cargo cult nigger behaviour
truke
you can give them legitimate info and they'll lash out like the subhumans they are, such as the anon who replied to you
>>
>>724672492
Pixels should not be an intentionally highlighted part of the texture
>>
>>724673140
>/v/'s entire knowledge of 3d modelling comes from a game of telephone
What exactly are you referring to
>>
>>724673451
Hey same fag, dont be a pussy, say something worth while or shut the fuck up.
>>
>>724628340
Its not the 90s anymore, 10k tris is not a lot, but don't go too crazy as there are diminishing returns
>>
Thats what a thought, bitch.
Useless.
>>
>>724673313
>you faggots chime in with info
I do that all the time but this isn't the thread for it. This is shitposting central.
>Chiming in like this isnt going to make you one of the cool dev
Who the fuck wants to be one of the "cool devs" or whatever category while anonymous, I'm literally fucking anonymous
>>
File: stellar_blade_shapekeys.jpg (826 KB, 2486x1958)
826 KB
826 KB JPG
>>724662210
How would you rate this and >>724672446 topology? I'm learning UE4 and really curious as the game overall technical side impresses me.
>>
File: STARMIE_GS.mp4 (1.87 MB, 998x624)
1.87 MB
1.87 MB MP4
>>
>>724672446
Is that Stellarblade? I can't really tell by looking at the model. But the filename has "SB" in it.
>>
>>724674005
Oh, nvm, it is Stellarblade
>100 shape keys
That's what I'm afraid of. I don't want to make a hundred shape keys to get good animations.
>>
>>724674214
Yeah.
>>
File: 1409885400112.png (12 KB, 510x546)
12 KB
12 KB PNG
>>724628340
if it's not FULLY MODELED I'm not interested
>>
>>724674379
> >100 shape keys
54 actually, and most of them can be mirrored.
>>
>>724650254
show us your bug



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.