Image not available

1920x1080

1701839593756750.webm

🧵 Untitled Thread

Anonymous No. 966268

How did they do this? Are these even haircards?

Anonymous No. 966285

>>966268
Brother this is old fucking tech. We've had this since the fucking nvidia 7000 series decades ago.

Plus chances are they tone this shit way the fuck down when this game ships.

Anonymous No. 966287

>>966285
Furthermore mirrors edge did this as well. You may wanna look at that as well as the witcher.

Anonymous No. 966291

>>966268
Strand based hair have been in blender as new hair curves for a while.
You can also use them in Unreal and Unity.

Anonymous No. 966305

>>966268
It's the sort of hair system that made maxed out Tomb Raider/Witcher 3 run like shit.

And it will probably still run like shit.

Anonymous No. 966306

>>966268
It's standard software that's pretty much everywhere. Bootleg Uncharted from 2013 had it as a big tech gimmick. Ran like shit back then.

Anonymous No. 966333

>>966291
Blender never had it until 2015, Maya and others had it. FF7 advent children has gameplay of it.

https://www.youtube.com/watch?v=wut2am39z-c&t=135

This is why blender will never become useful in any industry.

Anonymous No. 966336

>hair clipping through the shoulder
It's the same shitty RDR2 hair system

Anonymous No. 966346

>>966268
this is the longest hair you'll see in the game. western devs can't handle actual long hair.

Anonymous No. 966360

>>966336
It doesn't clip through.
Look closer.
>>966268
How do you know it's real time?

Image not available

762x572

Screenshot_2023-1....jpg

Anonymous No. 966364

>>966360

Anonymous No. 966366

>>966364
>well ackshually
Details like don't mean shit if it's during animation.
Game engine collision usually doesn't operate on meshes but on capsules that approximate the character mesh.
So what you see here is the discrepancy between a collision capsule and the rigged mesh.

Anonymous No. 966371

>>966366
>operate on capsules
That's historically true, but Bullet Physics and others have some options to do convex decomposition on the fly on animate meshes.
And follow that with convex/convex collision instead of point/capsule.
Certainly the PS5 (assuming they even exists because I've never actually seen one) has enough compute power to do that.

Anonymous No. 966375

If you want to try some state-of-the-art physics stuff:
https://github.com/bulletphysics/bullet3
You may have to compile it on your own and some stuff is experimental, but there's especially a demo of a robot hand grasping a stack of gelatine cubes that runs in real time on a modest computer. They also working on real time FEM simulations.

Basically don't lose faith because we're on track to get our own fully functioning waifus relatively soon and it's not a scam like AI.

Anonymous No. 966376

>>966371
I think you're confusing static meshes with rigged meshes.
Deformation makes mesh to mesh collision too slow. They can attach a tapered capsule onto each bone, with proper placement and dimensions these can approximate rigged character meshes with reasonable accuracy.
That's the usual workflow for Unreal and Unity afik and I doubt rockstar is doing something different.

Anonymous No. 966377

>>966376
That's historically true and you may be right. I just wanted to tell you that there have been serious advances in Bullet, which is Sony basically.

Also, you can try the Softbody support in UPBGE or even Godot (before a certain version), which is Bullet. It's somewhat incomplete, somewhat hard to control and nobody is working on it, but it does softbody/softbody collision without intermediate shapes.

The main problem is that Softbody support in Bullet is CPU-only at this time, so there is an upper bound of complexity/precision you can get.

Image not available

1442x837

cluster_collisions.jpg

Anonymous No. 966383

It's been a while since you could do interesting things even in UPBGE. Unfortunately there are a lot of bugs and missing code to make it useful in UPBGE, but I would be disappointed if Rockstar didn't have the balls to properly incorporate some of that stuff in GTA.

The Cyberpunk team kinda-sorta promised it during development, but eventually they didn't do it.

Anonymous No. 966386

For Godot you can take a look at this:
https://www.youtube.com/watch?v=xxLJjCpI58M
Which is how I believe they've made the hair in OP >>966268 and how they'll be making skirts and other dress items that respond to wind for example.
That's just my speculation but I hope they're going down that route. If not FEM simulations, which are starting to become available and my be even better.

Anonymous No. 966390

>>966268
I believe so, it's a hair mesh skinned with the vertices weighted to bones that are then solved thru physics simulation. 'Verlet integration' cloth can do these sort of things but setting it up is tedious. Looks awesome, best hair I ever seen in a videogame, knowing it's Rockstar it'll likely be performant not some gimmick that tanks your fps like other attempts at hair physics.

I've spent several weels attempting to build realtime hair system like this few years ago but had to move on because I was needed elsewhere, I'm deeply impressed with what I see here knowing what goes into it.

Anonymous No. 966398

>>966364
they don't clip, blonde hair can look like this when reflecting sunlight
t. dark blonde guy who has a few pics where it seems I don't have hair where the sun reflects its light

Anonymous No. 966407

>>966398
Why are you all trying to analyze shit tier bitrate video screen caps

Image not available

622x594

47542.jpg

Anonymous No. 966414

>>966285

Anonymous No. 966416

heh

Anonymous No. 966419

>>966268
>that god awful dithering on all the hair in the scene
Why the fuck do R* devs think that shit is acceptable. There's been shitty dithering in these games since GTAV. It's fucking horrible. Dithering isn't a solution.

Anonymous No. 966420

>>966419
brother games have been visually fucked since we left dx9 and went back to sega saturn checkerboarding.

Image not available

500x497

exde.png

Anonymous No. 966422

>>966414
'Ckin' hell

Anonymous No. 966433

>>966407
because I watched the original video in 4K

Anonymous No. 966450

>>966433
I believe you but none of that is here.

Anonymous No. 966453

>>966268
Not only is it some mid onlyfans girl looking 3d model that they chose to display. The hair doesn't even move like hair, no friction, no springiness. This looks bad.

Anonymous No. 966454

It looks like they are applying some kind of machine learning tonemapping to each frame look like a photo, but I have serious doubts that people are into playing these types of games anymore. Its just boring shit, just like cyberpunk.

Anonymous No. 966459

>>966268
the final movement she does with her right arm annoys the fuck out of me. after she puts it on her hip it kind of springs back like she's made lf rubber. ruins the whole thing.

Image not available

1280x720

maxresdefault.jpg

Anonymous No. 966489

>>966419
You mean this?

Anonymous No. 966493

>>966489
Nah, it's something about the RAGE engine specifically that has super fucking awful dithering. That example you posted is still pretty bad though.
It used to piss me off to no end when I played GTAV on launch on the 360. Literally everything was checkerboards.
The 30fps cap was bad enough but that dithering was next level horrible. Unfortunately it only launched on 360/PS3, so you had to wait for a "proper" release on PC, but the dithering still showed up to a lesser degree.
Looks like it's still present in the OP footage.

Anonymous No. 967642

>>966268
Looks disgusting.
https://www.youtube.com/watch?v=hVQax6olpUQ

Anonymous No. 967645

>>967642
I can literally see the framerate drop at 0:30
Why even post this like it's a dunk on op vid, this clearly isn't game ready retard

Anonymous No. 967653

>>966268
going frame by frame they are clearly hair cards
there is nothing impressive about the rendering, shit dithering and AA
what does stand out is the number of bones per strand, but i assume they made some in house solution that does baked sorting instead of self collision and other gimmicks to allow it to be driven with as few points as possible

Anonymous No. 967654

>>967645
>8 years ago

shhhhhh only dreams now

Anonymous No. 967655

>>967645
bunch of games used hairworks
its actually not that expensive, like 20% frame cost top, but so far they werent used for anything useful so i never found point in turning them on in Metro or other games to have monsters i dont even look at up close have tiny hairs on their balls

Anonymous No. 967656

>>967655
>but so far they werent used for anything useful so i never found point in turning them on
you never played the game

Anonymous No. 967658

>>967656
i play roblox with my brother a lot

Anonymous No. 967661

The animation is odd

Anonymous No. 967662

>>967661
That's because video game animations have no dynamics to them, they just interpolate from one canned animation to another.

Image not available

1280x720

gta.webm

Anonymous No. 967682

>>967661
>The animation is odd
Hold my beer

Anonymous No. 967703

>>967682
Now it looks worse, do better retard

Anonymous No. 967710

Why simulate real life when real life sucks?

Anonymous No. 968411

>>967661

It's not even just the animation, it's the action itself. Why is she doing that? She's just standing alone at a party and randomly strikes this pose for no reason. Is it the vaccine?

Anonymous No. 968512

>>967682
>the boobs in the background aren't rendered as well

Anonymous No. 968520

>>968411
Because she's aware of the camera behind her and wanna show off what she was up to influencing from that rooftop by striking the pose for the viewers.
Didn't you notice the recurring theme of all the self-obsessed people documenting their every act filming with their phone thruout the entire trailer?

GTA has increasingly become this dark mirror of contemporary culture deliberately designed to provide us with some perspective of current ongoings.
Rockstar's writers has a rich history of lampooning social media in particular so that aspect is prob gonna amp up in VI to be more in the forefront.

Anonymous No. 968556

>>967682
basically the chest has no dynamics to it. No jiggle, no bounce, no relaxation or flexion. No nothing. It looks fake.. And you could say that for the rest of the body as well. It looks last gen tbqhwyf.

Anonymous No. 968564

>>968556
Because it's not a coomer game, it's meant to appeal to both male and female gamers.
Sexy is therefore ok since that caters to everyone as having physical beauty is part of female power fantasy.
but soon as you start heading into borderline pornographic territory with it you begin alienating people.

>INB4 but jiggle = realism.

Yeah, but any solution for breast jiggle that runs realtime on a large number of characters will tend to look uncanny.
It'll either end up so stiff it wasn't worth the effort or so pronounced it looks fake and thirsty.

Could a tit-rigging wizard come up with a system that is both fast and naturalistic? Absolutely, but realize how it's a difficult spot to hire for.

Anonymous No. 968566

>>968564
>Could a tit-rigging wizard come up with a system that is both fast and naturalistic? Absolutely, but realize how it's a difficult spot to hire for.

its not hard to do now. You can do it in any engine. You use a machine learning library extracting the differences from a muscle sim to Linear Blend Skin and run it on xpu on characters close to the camera. There are already free and paid tools that can do this. I have done this personally and the results are stunning.

Anonymous No. 968571

>>968566
I'm sure you've done all that in your head and it works perfectly there but if you want to know how to do it in reality:
https://github.com/bulletphysics/bullet3
clone, compile, run the demos, read the papers about the softbody simulations and just go from there. It's been out for a while but has never been exposed in the GUI of your Blender or whatever for obvious reasons.

Anonymous No. 968573

I'm sorry I've mentioned Bullet so many times
>>966371 >>966375 >>966377 >>966383 >>966386 >>968571 but it really is just that.

Anonymous No. 968574

>>968571
what the hell are you talking about. You have no idea what im taking about

Anonymous No. 968580

>>968574
I do actually. You're talking about the Houdini 20 keynote but that's meant for offline movie production.
What's used for games is something called "tetrahedralization" or "3D Delaunay". It's a process that allows you to run collisions and FEM simulations on a model very quickly.
You can play with it in Houdini (because it's available in there somewhere) or you can do so via a physics library. Or if you're crazy enough, you can even do it from scratch:
https://www.youtube.com/@blackedoutk/videos

Anonymous No. 968581

>>968580
No. You can do it with ZivaRT with machine learning extracting the differences from muscle sim (created in any package) to LBS

Anonymous No. 968584

>>968581
And how would you do that for a ragdoll for which you can't predict the movement in advance a videogame in realtime?

Anonymous No. 968586

>>968584
There are notes about that very issue in the recommended workflow guides.

Anonymous No. 968587

>>968586
And what those notes say?

Anonymous No. 968588

>>968587
>generate a wiggle ROM
>add more training data
>use smaller patches

Anonymous No. 968594

>>968588
And how do you deal with collisions and self-collisions if all you have is a glorified set of shape keys?

Anonymous No. 968596

Also, do those notes tell you how to pirate the commercial software and obtain all the training data that you need for the machine learning stage?

Anonymous No. 968597

>>968594
those >>968588 are the steps you need to take for novel poses outside of the training data

>>968596
if you have the pirate mentality you will never ever make it.

YOU are responsible for generating the training data by caching out various combinations of deformers / FEM / Vellum simulations that dont run in realtime

Anonymous No. 968598

>>968588
This sounds so very trivial to implement anon. Why don't you contact Rockstar and tell them all about your machine-learning titties and bless us with this simple solution you've cooked up?

🗑️ Anonymous No. 968600

>>968598
I dont play movie games and i hate rockstar

Anonymous No. 968601

>>968600
So you just happen to have these really strong opinions about accuracy in depictions of fictional breast tissue?
How come you hate Rockstar if you don't even play games?

Anonymous No. 968602

>>968601
I've said too much already.

>How come you hate Rockstar if you don't even play games?
I don't play moviegames. I only occasionally play Real games (fgc, from software, resident evil)

Anonymous No. 968603

>>968602
>I don't play moviegames.

Suppose you mean cut-scene heavy games? Yeah I don't care for that aspect either I think exposé in gaming is best kept snappy and short to just set you up for the actual playable segments that should be 99% of what you do playing a game. With rockstar titles the worlds they create are so rich in detail and facilitate free-roam sandbox gameplay that it end up being 99% gameplay for me despite them hours of cutscene exposé.

Shame you won't provide your ML-titties tho, I have thousands of hours on my female protagonist GTAO account and could get a lot of joy out of them knockers.

Anonymous No. 969124

>>966268
that's xgen virtual hair, the game is gonna look wayyyyyyyyyy worse than 'at.

Image not available

1080x607

Screenshot_202401....png

🗑️ Anonymous No. 969862

>>966268
its like she has a metal head and her hand is a magnet, good thing is gamers will be too busy counting every pore on her nose
>>967682
AI titties!

Anonymous No. 969878

Can someone explain how strand based hair work on a technical level in game engines?
Naively I'd expect that each hair has to be rendered as a line segment. In that case if each hair has say ~30 segments and there's 1000 hair that's 30k primitives that have to be drawn.
Surely that's not the case or is it?

Anonymous No. 969885

>>969878
https://www.google.com/search?q=hair+cards
And softbody physics.

Anonymous No. 969886

>>969885
I specifically asked about strand based hair, not hair cards.
If your claim is engines convert stands to cards automatically provide evidence for that.
Else learn to read.

Anonymous No. 969887

>>969886
Strand by strand is offline only and try not to get too emotional about hair or fur.

Anonymous No. 969889

>>969887
https://docs.unrealengine.com/5.3/en-US/hair-rendering-in-unreal-engine/
>Traditionally, hair that has been created for use in real-time engines has been created using card-based techniques, or other similar approximation. Card based workflows use a large number of flat sheets (or cards) to give the approximate shape, and motion, of a much larger number of individual hairs. Using the Groom system and tools of Unreal Engine enables you to render each individual strand of hair, which significantly improves visual fidelity of simulated hair in real time.
The UE docs claim otherwise.

Anonymous No. 969890

>>969886
What we see here is haircards tho, look carefully and you see how lumps of hair always move together, it still looks convincing as long as the hair mesh is done right. And hair meshes in AAA games are ridiculously dense these days, go dissect some and you'll see.

I don't know what optimization the systems you talk about that are truly strand based use but I know they tank performance so it's not really that impressive to me as they're unusable for production for the foreseeable future.

Hybrid approaches like these looks great and if they perform as depicted here that will be awesome.
Rockstar's trailers historically show stuff highly representative of in-engine footage, not this fake dressed-up stuff that get degraded towards launch.

Anonymous No. 969892

>>969889
If it's a new feature, I didn't know about that. It's probably shader based. If it works in real time on a less than $10k computer, it could bring Ton Roosendaal's beard simulation to even higher levels and I'm sure it'll be added to Geometry Nodes with the highest priority.

Image not available

1446x854

tifaHair.jpg

Anonymous No. 969893

>>969890
>hair meshes in AAA games are ridiculously dense these days, go dissect some and you'll see

Case and point Tifa's model from FF7 remake. 44 thousands triangles for her hear, 1/3 of the polygons for the character model is hair.

Anonymous No. 969898

>>969892
>It's probably shader based.
Thanks, that's what I wanted.
Perhaps it's using raymarching signed distance fields? If it's possible to use arbitrary and non-convex shapes as raymarching volumes I think that could work.
This of course would not enable individual strands for physics simulation but from what I've seen that's also not really happening.
>>969890
>Hybrid approaches like these looks great and if they perform as depicted here that will be awesome.
There's certainly some trickery involved and I'm in no delusion that they actually "rendering and simulating every hair".
I could conceivably figure it out myself if I read through and try to understand the corresponding UE source code. But then I'm lazy.

Anonymous No. 969906

>>969898
Besides the visual aspect, there's also the cost of collisions and self-collisions. With hair cards is possible. On a single hair by basis I don't think it's going to become possible in our lifetime.

Anonymous No. 969908

I've finished reading the Unreal engine thing.
It says: "For example, you should expect a 30Hz or higher framerate with an RTX-2080Ti for a human-like groom at 1080p."
No shaders, hair by hair with physics. Kind of impressive.

Anonymous No. 969909

>>969908
> 30Hz. Kind of impressive.

And this children is why video games never ran smooth no more.

Anonymous No. 969911

>>969909
just use frame gen bro
make your game a smeary mess bro it's fine bro

Anonymous No. 969997

>>969908
>No shaders, hair by hair with physics. Kind of impressive.
It says they render each strand. Nothing in these descriptions is unintentional so I'm pretty sure it's safe to say they don't simulate each one.
Why is that important? Because it leaves open the possibility to render it faster/differently than a line segment for each section of a single hair.
>>969911
Lumen already more or less requires TSR already so there's that.
On the upside if there's enough performance headroom available (4090) the picture clears up somewhat.

Anonymous No. 970877

Since this is now the general hair thread I ask here about this idea I had:
Could using raymarching signed distance fields be practical for short hair rendering?
(For those who don't know what that means look on shadertoy)
The idea is to offset the rigged mesh buy a fixed offset along the normal direction and use the resulting two meshes (original and inflated) as ray marching volume.
The one potential pitfall I see is how to compute intersection with the volume as this falls into the "arbitrary mesh intersection" category which is no bueno.
But perhaps someone has an idea.

Anyway the neat thing is if this works it might be more performant than the unreal thing from above.

Anonymous No. 970886

>>970877
To compute the intersection between a mesh and an SDF is easy but once you know that there's an intersection, what you're going to do?
To bend an SDF is an inexact operation because it changes the metric. It can be done but only to an extent.
But you're welcome to try.

Anonymous No. 970891

>>970886
The general idea is to map the xy plane of the coordinate system onto the character mesh surface and the z axis towards it's normal.
The SDF parameters could be encoded in textures. (for every UV island)
A normal map for hair direction, and one additional texture for hair thickness, length and possible curvature and rotation along z.
Just use every texel with more than 0 length as a point where a hair grows.
The base SDF could be either a line segment or circular segment and the thickness can be done by using an offset for the 0 value of the SDF.

>To bend an SDF is an inexact operation because it changes the metric. It can be done but only to an extent.
I'm not quite sure I understand. Each face would correspond to a volume that's the prism made by extruding it along it's normal.
If you mean the gaps between these I was thinking I'd simply merge the vertices from the extrusion onto the median vector so the mesh is a consistent shell. This should work for convex and concave geometry.
Sure above the boundaries for each face the hairs would have a "kink" but perhaps that's not really noticeable.

Anonymous No. 970901

>>970891
You're speaking hypothetically and I can't comment on that.

Anonymous No. 970914

>>967682
you're hired
but no srsly, why don't they release vidya that pulls a big fuck you to all the vidyia moralfagging, with blood gore cruelty sex, nudism, brutality, etc