🧵 Arcsys topology question
Anonymous at Fri, 2 Aug 2024 22:55:49 UTC No. 991417
I know why the fans around the lips exists but why are there so many cuts on the planes all through the head?
Usually when I see people make anime models, the topology is just regular planes.
Anonymous at Fri, 2 Aug 2024 23:10:05 UTC No. 991419
>>991417
afaik all game engines triangulate quads at some point, so ripping the model from the game gives you all tris
Anonymous at Sat, 3 Aug 2024 00:26:13 UTC No. 991421
>>991417
either to save on tris like in the area around the jaw, or to shade the model in a particular way, like to fake the Rembrandt triangle thingy below the eye.
Anonymous at Sat, 3 Aug 2024 02:28:53 UTC No. 991426
>>991419
That's right. Quads aren't game engine friendly, so the models get triangulated. That's why modellers for games are less concerned about keeping their models all quads. Since it's getting triangulated anyway, you might as well use triangles to fine tune your model.
Anonymous at Sat, 3 Aug 2024 05:20:10 UTC No. 991438
>>991417
Shaders uses the area of faces and their normals to calculate how a mesh should be lit. Their models probably uses standard face topology, except the loops around the lips are bigger because it also means bigger lighting changes from normal editing. The current look is what >>991419 said.
Anonymous at Sat, 3 Aug 2024 14:54:03 UTC No. 991456
As the other anon said, models exported from game engines are already triangulated which is why they look like that. To explain the weird cuts near the eyes, those are made so ArcSys can control the shading to make it more anime-like. This process is known as 'custom normals' though the process is moving away from doing it via making cuts in the topology to doing it via an SDF map because doing it via topology means it breaks the moment it's deformed. Doing it via this SDF method means you can deform the face all you want but still keep the shading how you want it. The closest analogue to an ArcSys game that's made use of this new method is Hi-Fi RUSH though Persona 3 Reload and miHoYo games have also taken this approach despite not going for the ArcSys anime style.
>>991438
No that's not it. It's because ArcSys characters have very exaggerated and expressive anime expressions. The fanning around the mouth makes it easier to deform and contort the mouth into these exaggerated expressions. That's all.
Anonymous at Sat, 3 Aug 2024 16:00:48 UTC No. 991461
>>991456
>>991426
>Game engine
It's not a game engine thing, it's a 3d rendering thing. A polygon has to be coplanar (ie flat) to be rendered and the only polygon garanteed to be coplanar in 3d is a triangle (because it has only three points so they define a single plane) so everything is triangulated when it gets sent to a GPU or renderer, even if it's behind the scenes. Due to finite precision that's true even for polygons that should be flat, let alone for the organic curvy shit you get with characters
Anonymous at Sat, 3 Aug 2024 16:03:36 UTC No. 991462
>>991456
worth Arcsys don't use SDF maps, just custom normals (but if it's well done it's functionally equivalent, a SDF map is just more intuitive to tweak because you straight up draw what you want it to look like in the end
Anonymous at Sat, 3 Aug 2024 16:12:24 UTC No. 991466
>>991462
I don't know if ASW plan to move over to using SDF maps eventually. Big issue with using an SDF map to control lighting is that it's faux lighting at the end of the day (limited to Z-rotation) compared to proper custom vertex normals. Also the shading in the area where the face connects with the body won't be consistent. You can get away with it if you use your lights wisely but ASW already have a system in place to tweak normals per frame of animation to fix deformation artifacts. I think if you want a 'fire and forget' way of getting good face lighting then this SDF method is the way but if you want to be as anal as ASW and control every single facet of every single frame then custom vertex normals is the play.
Anonymous at Sat, 3 Aug 2024 16:14:57 UTC No. 991468
Why are so many fuckers trying to go out of their way to make 3D look like something it's not. Why not develop an artstyle that fits the media instead of trying to fit the square peg thru the round hole with horrific looking results save at the very pinnacle of the craft where it just look slightly off-putting.
You wanna do 2D artstyles, why not just do 2D?
Anonymous at Sat, 3 Aug 2024 16:31:14 UTC No. 991469
>>991419
Game engines doesn't triangulate the geometry, they just throw away the isoline display used for humans to more easily read and manipulate
surfaces as well as keeping them ready for subdivision by telling which edges the subdivision is to happen across.
The surfaces are always triangled wheter you are doing tris, quads or N-gons, your software can be told to display the triangulation for you at any time.
>>991438
Shaders don't use 'the area of faces' to decide normals, each vertex has at least 1 normal vector that tells what direction is considered up along the
adjacent edges from that vertex. to decide how the area is to shade along a face you interpolate what the normal is at the point of the pixel being drawn
by taking the average of the three vertices of the triangle you're on weighted by the distance to each vertex.
If all edges connected to a vertex is set to render smooth it will have one normal but if the edges are shading hard the vertex associated with it will have more than one normal up to a number depending on how many hard edges meet at that point.
Anonymous at Sat, 3 Aug 2024 16:32:37 UTC No. 991470
>>991468
>Why are so many fuckers trying to go out of their way to make 3D look like something it's not. Why not develop an artstyle that fits the media instead of trying to fit the square peg thru the round hole
Because it's a fun challenge. Photorealism has been solved. 3D that could pass as 2D hasn't.
Anonymous at Sat, 3 Aug 2024 16:39:48 UTC No. 991473
>>991466
i mean wouldn´t you just be able to do the same thing but by painting directly on the SDF map?
https://youtu.be/WyxPN2HRaFk?si=LRr
the thing is 90% of the time you only really need Z-rotation since that´s where you get all the "anime shadow patterns", for anything else you could just turn it off and rely on the default normals, maybe do some deformation fuckery that only looks good from the right angle since it´s probably a one time thing anyways
Anonymous at Sat, 3 Aug 2024 16:49:12 UTC No. 991475
>>991470
Yeah but why not do something new, like develop NPR art styles that are meant for 3D? You're trying to do something inherently compromised in forcing an aesthetic onto a usecase different from what it was originally developed for, and the community has been failing at it for ~20 years now.
Anonymous at Sat, 3 Aug 2024 17:31:30 UTC No. 991482
>>991480
saying it uses the area to decide normals implies that the area of the triangle determine what the normals are. This is completely incorrect and indicative of a fundamental misunderstanding of what's going on.
The normals are whatever they are set to be at the vertices however and the area of the triangle is completely irrelevant to what they are. Your triangle can be however big or small, accute or obtuse and have the normals point in whatever direction you want.
Anonymous at Sat, 3 Aug 2024 17:31:36 UTC No. 991483
>>991475
>why not do something new, like develop NPR art styles that are meant for 3D?
Plenty of games do, you're just being difficult by bringing that up in a topic about anime models
Anonymous at Sat, 3 Aug 2024 17:37:22 UTC No. 991484
>>991469
>Shaders don't use 'the area of faces' to decide normals, each vertex has at least 1 normal vector that tells what direction is considered up along the adjacent edges from that vertex.
The increased use of compute shader skinning instead of vertex shader skinning has also brought fully dynamic compute shader normals into the mix, in which case weighting by face area is indeed the usual way it's done.
It's got the advantage that for skinned models normals were actually wrong, because the calculation usually used for them was a shortcut that avoids expensive inverse transforms but was "good enough", and now you can fully compute a normal based on the surrounding faces rather than transform a premade one.
Anonymous at Sat, 3 Aug 2024 17:39:59 UTC No. 991487
>>991483
>you're just being difficult by bringing that up in a topic about anime models
Nah bro, I'm a reformed weeb who tried to make anime myself in the early 2000's using the 'illustrate!' plugin employed to render the tachikomas and other assets in GitS SAC.
I'm just here as someone who have once traversed these planes standing like a haggard boomer in a cloak telling you how the princess is in another castle type thing.
Anonymous at Sat, 3 Aug 2024 17:41:32 UTC No. 991489
>>991482
>his is completely incorrect and indicative of a fundamental misunderstanding of what's going on.
No, it is in fact an existing method and does give better results than older methods.
This is the 2007 article that started it but 17 years later all modeling programs implement it
http://www.bytehazard.com/articles/
Anonymous at Sat, 3 Aug 2024 17:55:26 UTC No. 991491
>>991489
That article describes an algorithm that set the normals at each vertex according to the surface area of adjacent faces.
It's still not the same as saying that the area of faces is what decides normal. It's part of what set defaults if you use this algorithm
but you can ofcourse then edit those normals to whatever you want them to be because the rendered normal is a vertex property not a face property.
I suppose you can write a custom format that do store the normal per face but that is def non-standard.
I don't know if compute shaders utilize such schematics as I never written any so can't comment on that part.
But the problem of talking about normals like that is how I could sit down and write an algorithm that set my normals
based on the position the moons of saturn and then state that saturns moons decides normals and I mean technically I would not be wrong.
Anonymous at Sat, 3 Aug 2024 18:21:09 UTC No. 991495
>>991480
I've always wondered how arcsys' 3D modeling technique works for characters like Kaido or Blackbeard or Kizaru.
I'd imagine it would be easy for characters like Zoro or Nami because they're relatively simple designs, but I can't imagine it being easy for someone realistic-looking like Kizaru or inhuman like Kaido and still having them be as expressive as they are in the anime.
Anonymous at Sat, 3 Aug 2024 20:42:35 UTC No. 991501
Are there any good tutorials for Arcsys' method on Youtube? I found two but they're both for generic anime girls. So far I've been using them and that Arcsys Dev' lecture video.
Anonymous at Sat, 3 Aug 2024 21:26:51 UTC No. 991507
>>991456
I'm interested in this SDF map because normal editing is a fucking pain
Anonymous at Sun, 4 Aug 2024 21:46:27 UTC No. 991576
>>991507
Here's a good blender setup
https://www.youtube.com/watch?v=x-K
Anonymous at Sun, 4 Aug 2024 22:53:03 UTC No. 991580
>>991495
They straight up swap meshes (shown in the original 4gamer Xrd article) or add extra meshes like the brow furrowing in spiderverse (shown in the cgworld revelator article).
Anonymous at Tue, 6 Aug 2024 01:48:43 UTC No. 991637
>>991456
>>991417
How do these games get the outline effect? I find it difficult to believe they're using the reverse hull method. Because that would be very expensive.
Anonymous at Tue, 6 Aug 2024 04:33:38 UTC No. 991644
>>991637
Hulls indeed. Z prepass means it really isn't much of an expense
Anonymous at Tue, 6 Aug 2024 04:48:25 UTC No. 991648
>>991644
>Z prepass
What does that mean?
Anonymous at Tue, 6 Aug 2024 05:18:33 UTC No. 991650
>>991648
Scebe is first rendered as depth only, without any shader, then rendered a second time with shaders and depth test set to equality. Means you transform geometry twice but shader costs is only paid on visible pixels, so in the hull case you only pay for the outline pixels.
Anonymous at Tue, 13 Aug 2024 19:50:09 UTC No. 992104
>>991466
>ASW already have a system in place to tweak normals per frame of animation to fix deformation artifacts
Can you explain this system or link to where I can read more about it please?
Anonymous at Tue, 13 Aug 2024 20:13:19 UTC No. 992107
>>991468
I wanna make a 3D game but I want it to look like anime.
Anonymous at Tue, 13 Aug 2024 23:31:13 UTC No. 992115
>>992104
That anon's wrong, according to the 4gamer article translated on polycount, they don't adjust the normals per frame, it's a single per-character light they can move around per frame that they use to tweak lighting. But they really only mostly use it for cinematic moves.
Anonymous at Wed, 14 Aug 2024 00:40:43 UTC No. 992118
>>992115
It's not mentioned in the 4Gamer article retard. You're talking about something completely different. ArcSys (and CC2) use a custom vertex shader in Unreal to manipulate the face normals per frame of animation to fix the fucked up deformation artifacts that appear once the face bones start moving. Go and download UE Viewer, you can see it for yourself.
Anonymous at Wed, 14 Aug 2024 00:45:22 UTC No. 992119
>>992118
Also you can't just data transfer correct normals from a proxy mesh onto the model in Unreal like you can do in blender which is why a custom vertex shader is necessary, but the SDF approach completely side-steps the need for either of these.
Anonymous at Fri, 23 Aug 2024 01:43:33 UTC No. 992644
A couple of questions I have about the arcsys method
>1
Is there a good guide on how to make a face model? So far all I have is this
https://www.youtube.com/watch?v=upt
and it helped me for a bit, but it's for generic anime girls so it's kind of hard to apply that to other anime - which I've done, but I want an easier method that makes my topology look more polished
>2
I have the addon for square UVs, but before I make an entire body, I just want to try a simple head model first. The problem is that the map looks all messed up when I unwrap the head. Any suggestions? Should I add seams?
>3
how do the lips and fans around the mouth actually work? I have both PDFs from Arcsys' lecture about them but they don't go into detail on what I need to do. The video I linked helps, but not for bigger lips, nor the fact that some of the triangles have 3 lines, or the space of vertices below the bottom lip
Anonymous at Sun, 25 Aug 2024 21:11:58 UTC No. 992794
>>992644
Basically see the whole moth as an independant piece to be deformed and moved anywhere on the lower face, much like a cel would be. The role of the fan is to be as unobtrusive a fill around the mouth as possible.