Does anyone know what Ni No Kuni did "wrong" here? The cel graphics look very flat 3D and un-anime compared to GG Xrd.
https://www.youtube.com/watch?v=RuallI8uFlI
https://www.youtube.com/watch?v=RuallI8uFlI
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
GDC : How Arc System achieved the cell shading in Guilty Gear Xrd
- Thread starter dosh
- Start date
Sunnington
Banned
The video explains how they did it in detail. It's actually quite complex. Each model used hundreds of bones to achieve the visual effects they wanted.
Level 5 animated the game according to the industry standard for cel-shaded animation. That is to say, they borrowed the artstyle and palette of Ghibli's works, but not their animation techniques.
This is why the presentation pointed out that you did not need a team of 3D animation experts. On the contrary, most of the look and feel come down to familiarity with 2D animation, something ArcSys has in spades.
This is why the presentation pointed out that you did not need a team of 3D animation experts. On the contrary, most of the look and feel come down to familiarity with 2D animation, something ArcSys has in spades.
I guess it means L5 needed more triangles than PS3/3DS could provide.
It's really, really not that simple. It's like asking why all photorealistic games on the PS4 don't look like inFamous: Second Son or The Order: 1886. Not the same techniques are used. Not the same level of devotion to the art style is present. Not the same talent or level of experience from the developers is involved.
Fun fact: The software they used, Softimage, was discontinued.
Huh, interesting.
Yeah, it's been kind of disheartening, because Softimage was way ahead of Maya and Max in a lot of ways.
Autodesk has been implementing a lot of Softimage features in the latest releses of Maya though. It's still missing some of the tools mentioned in the GDC video, but there's still hope.
I'd say the main difference is the use of normal orientation. It gives ASW a ton of control over the way light affects their models. Nino Ku Ni uses a more classic approach, that usually generates a lot of small inconsistencies.
Even in Processing Mode?
Because the game runs fine for me outside of using Quality Mode.
JordanN
Banned
So I just watched the video and one thing he brings up was about the normal map. He says the studio didn't use them because they wanted to maintain a clean look and so a normal map (which is dependent on texture resolution) could cause artifacts.
It's understandable that because there was a PS3 version, they were concerned with texture limit. However, if it was strictly focused on just PS4, would that have changed the studio's position?
It would make life more easier if they did find a working solution to using normal maps instead of having to manually tweek the vertex normals. Not only that but they would be able to do more with it by manipulating the normal maps in a shader or making their own custom maps.
Edit: I only brought up the detail map suggestion because I thought they were worried about textures breaking down when viewed close up (not to add anything that would mess with the artstyle).
It's understandable that because there was a PS3 version, they were concerned with texture limit. However, if it was strictly focused on just PS4, would that have changed the studio's position?
It would make life more easier if they did find a working solution to using normal maps instead of having to manually tweek the vertex normals. Not only that but they would be able to do more with it by manipulating the normal maps in a shader or making their own custom maps.
Edit: I only brought up the detail map suggestion because I thought they were worried about textures breaking down when viewed close up (not to add anything that would mess with the artstyle).
And it sounds bad. Like 64kbps MP3 bad.
The issue isn't maintaining a high density of high-frequency details across surfaces, it's maintaining cleanliness and clarity in how large-scale details are resolved. Detail textures solve a problem that doesn't even come up in this kind of art style.
Graphics Horse
Member
How would detail maps help make anything look clean? The type of look they are going for isn't benefited by adding noise, you'd just end up with grainy boundaries between tones.
Whenever you think "it shouldn't be that much work", you're falling prey to the dunning-kruger effect (ie "I'm not versed well enough in that domain to realize how much work it is"), especially since they do say in the video it took tons of effort.
Nah, relying on bigger maps is just short-term thinking. Using vertex data instead you have clean data, more compact, forever. And authoring the custom normals isn't any harder than authoring a normal map in the first place.
JordanN
Banned
A normal map doesn't have to be used for just noise. It can be used to correct/create better object shading. Because the artstyle isn't going for complexity, then I guess you can forget the detail map then.
In the video they actually mention that the fact that they were working on a 2D fighting game was an advantage, because the perspective is fixed and you have complete control of the lighting per character. Making something look 2D where the camera is rotating around the model would be way more difficult.
Also rotating the camera means that you couldn't maintain the low FPS on the characters without it looking jarring, and the lowered framerate was key to making it look really 2D as well.
There would be other issues to deal with too, like way more characters to animate in Ni No Kuni (taking 2 months to animate a single character with a limited set of moves. It wouldn't be feasable in an RPG with many more characters and lots of acting involved)
I'm not sure what you're suggesting.
The "corrections" they need are provided by their vertex normal data, essentially the same as a normal map but with different storage format and runtime filtering.
It's also by necessity unique to different parts of each object, which runs totally opposite to the concept of detail maps (which are all about taking advantage of redundancy across large areas of a surface).
JordanN
Banned
The models they showed had 40,000 polygons so that's what had me worried about how much work must goes into adjusting every vertex normal.
If that's actually an issue, it's more a tools issue than a method issue. For their purposes, in theory, I don't see why vertex normals should pose bigger authoring challenges than normal maps.
It doesn't work like that,most of the time they use proxy objects that transfer their normals to the final model. That, and Softimage has excellent normal editing tools (one of the many reasons it will be mourned).
Theonik
Member
In fact, doing it this way is significantly harder and labour intensive, especially for characters that need significant amounts of deform animations where they need to make multiple versions of each model altogether.
Where they benefit is in having the ability to inexpensively animate the camera and scaling to multiple resolutions.
Yeah most of the industry doesn't care for stylized games that much, when it would actually be less resource exhaustive than their current efforts.
And it would look more pretty and timeless when done right. Many CRPGs of old for example, still playable since they use handrawn sprite and background image. I really wished that more western developers that is not indie took this route instead of chasing for endless photorealism.
Yup, that, and Gator - I assume that's what they're using to transfer their normals. Although I swear I've seen a plug-in specifically developed to do that (transfering normals from one object to another) in XSI.
Can't wait for it to hit PC if only to admire the technical work they put into it. They did an outstanding job. The best thing is that since it's all 3D it's gonna scale really well to higher resolutions 
Definitely gonna watch the video later.
Bloodstained though!
Definitely gonna watch the video later.
Bloodstained though!
Theonik
Member
Doing what they did is very costly and would be comparable if not worse than their current efforts. Not to mention a lot of their tricks are not really applicable if you are not making something like a fighting game. What they've done isn't exactly novel either but a very interesting application of existing techniques motivated by targeting a very specific goal.
Using normal maps means they would need to worry about producing UV coordinates that wrap the map across all polygons in an uniform way. It would take ridiculously high resolution maps and it would still not prevent staircase artifacts in certain areas, specially crevices and irregular surfaces. Texture compression would also introduce artifacts in normal maps containing mostly smooth low frequency detail. Normal maps also use 8 bits per component while vertex normals use 32 bits per component.
In short, it would be very hard to keep the shading smooth with normal maps, even on the PS4.
sixteen-bit
Member
The technique results in a product that does not transfer well to any genre expect fixed camera fighting games as well.
Their pipeline is full of really smart tricks that can seem a bit time consuming during pre-production, but nothing that prevents them from using it for, say, an rpg or a anything else really.
On the contrary, the way they build their characters ensures a lot of consistency in the rendering without having to worry about the angle from which you look at them.
sixteen-bit
Member
Really? In the Q&A section the ArcSy employee said it wouldn't transfer over to other genres, specifically those with 3D movement.
I don't think this is necessarily true. The technique can certainly be developed further and while the look that is indistinguishable to 2D animation might not transfer well if it can take you 90% of the way it might be enough for those other kinds of games.
the exact approach they used wouldn't transfer well, that's all he really said about it - but with workflow improvements, tools engineering and refining of the technique/theory there's no telling how far this might go. he doesn't strike me as an engineer, and while his initial creativity and lateral thinking might have been what made this possible in the first place it's up to tech people and engineers to refine and simplify the process in the future.
I think he means that in a full 3d environment, where you can move your character or the camera around freely, you would have the impression that the lighting on the models feels a bit limited, or that the light doesn't impact the characters as much as in a game like Nino Ku Ni for example. But it's only a problem if that doesn't suit your artistic direction.
What I mean is that it's probably a question of artistic choices rather than technical issues. I might be wrong though, but I don't really see what would prevent a developer to use this pipeline on an rpg for example (other then requiring a lot of work).
NormalFish
Banned
are there any screens of this game downsampled? aliasing is the only issue I have with the art style and I'd love to see how it looks without.
that and for some types of games any kind of perfect 2d -3d is just not possible. for example a 3rd person game with free camera. 2d animation relies alot on view dependent volume deformations and stylized lighting that doesn't respect physics - to generalize that into something that is indistinguishable to 2d animation in a free camera environment seems impossible to me without tons of time consuming artist authored tricks. For many games though, I'd argue if they get the shading right and work hard on the animations it's enough to get it to a point where it works.
Theonik
Member
No PC version so no.
Lighting itself is unlikely to be the problem, actually - their solution looks good under any angle. Animation is, however, as many deformations only make sense when viewed from the side or predefined camera angles. There's also other stuff which he didn't touch upon in the GDC presentation but did in the earlier 4gamer technical article: the models don't follow the standard laws of depth and perspective, and alterations are very 2d-specific (for example, each character has its own vanishing point).
My comment extends to fighting games. I would be surprised to see any game that isn't Arc System Works' adopt this method in the future.
FlashbladeNeo
Neo Member
I thought he said the change in lighting/atmosphere didn't lend itself well to a fighting game where you would need to be able to distinguish the move sets of each character clearly.
I just took his comment on agreeing it wouldn't work as a lack of being able to articulate the proper response in english.
I feel once you had the model and lighting done for a character, it could work for an RPG since the mood and atmosphere is more important than the characters precise actions.
In a [2d] fighting game clearly seeing the move set and frame data is important when trying to master the game. I think that was what he was getting at to why it wouldn't work for them in changing the lighting versus another genre which would be ok since most actions from a distance would be more basic (Fighting=20 different punches versus RPG = 1 punch)
Any game that wants to make use of their workflow would need a cap on the maximum amount of perspectives with which the player can view a character. Otherwise there would be no way to cover all the lighting cases to stamp out the visual "artifacts" (quotes because they're only artifacts in the context of 2D animation) of fully automated shading.
So, something like SMRPG could be done in this style. Or many old top-down or fixed perspective RPGs. Side-scrolling platformers would work too.
But nothing like Wind Waker or, oddly enough, Valkyria Chronicles.
So, something like SMRPG could be done in this style. Or many old top-down or fixed perspective RPGs. Side-scrolling platformers would work too.
But nothing like Wind Waker or, oddly enough, Valkyria Chronicles.
its hard to say how much of a problem it is. he said during the talk that each character has their own set of lights(no global lighting) that are animated to get the right shadows and highlights.
CamHostage
Member
But Ni No Kuni also has a controllable camera (or at least an oriented camera that follows characters in 3D space.) Part of the magic of Guilty Gear Xrd was that it did everything to make it seem like everything was 2D, so when it broke out of that dimension, it was like a magic trick that overloaded your mind.
(I felt a little similarly about the graphics in Gears of War. It's for sure a rich and expensively produced game that looks awesome in regular action, but the real convincer for me was how effects kicked up an extra notch in the roadie run, how the blur and shake and feel of elements whisking by felt that much more real and exciting.)
Not every game should look like a 2D cartoon (and although traditional cel-shaded animation could benefit from some of the improvements developed in GG Xrd, not everything is applicable and not everything needs to be changed from the older methods in order to make a game look good.) So let's not jump to the conclusion that everybody else working with variants of cel-shading techniques are and have been doing everything wrong. But I do hope more developers look at the accomplishments here and are inspired.
Maybe not games with 3D movement, but I could see this working in any traditionally 2D game that you'd want to have a more dynamic camera while still looking hand drawn.
Shmups, beat-em-ups, 2d platformers, metroidvania, etc. could benefit.
Plus, I could see the lateral thinking-approach being carried over to 3D games for more unique art styles. As Yoshi's Woolly World has shown, many developers are already crushing it with alternative graphical art direction.
Yup, it's one directional light per character and it looks fine, it's not view dependent, ie you can rotate around fine and it always looks good. That's because the normals are tweaked to look universally good (so they don't need to adjust them on each animation). It's nowhere like the character deformations, which look wrong when the angle changes (search the 4gamer article for the elongated May so see what it's about).
Similar threads
- Poll