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



Thread archived.
You cannot reply anymore.




File: 1533983200197.png (563 KB, 1108x388)
563 KB
563 KB PNG
Why does one of these models look good, and one bad? Their topology seems about the same quality. I'm don't know much about 3D so I don't understand why the shading is so fucked up on the one on the right.

References:
https://www.youtube.com/watch?v=ioX5EwyEuTQ
https://www.youtube.com/watch?v=MNb50Xvkmrs
>>
>>637356
They look both good to me, left eyes maybe a bit high
>>
Lighting on the right is fucked
>>
Lighting and texturing. There's certain methods needed in achieving a good 2D look from 3D, and objectively neither have really done so.
The first one on the left just went completely unlit, i.e. there's no lighting in the scene and the model is without shadows and highlights. It's a good way to hide lighting issues that may arise with trying to get a 2D look and the lighting details you see have pretty much been drawn onto the texture. More gradients could have been used in her texturing to make it look more lit up, though, so now it currently looks quite plain and boring.
The one on the right is also pretty similar, except it's not going unlit and rather some sort of "general lighting" that still casts some shadows, you can see them most prominently along the mouth. The texturing for her is much more detailed with a lot of faux ambient occlusion and simple gradients going on.
However, there's a lot more work involved if you want to have a character that shades correctly like a 2D drawing, and whoever created her went maybe about 50% of the way. The biggest of which is normal manipulation, bruteforcing vertex normals to align a certain way to "hack" lighting. I personally don't quite understand it myself, but it creates very useful effects even outside of this 3D to 2D thing. Pic related. It's also how some VRchat animu avatars look straight out of a chinese cartoon when all the rest look terrible.
In short, whoever made the one on the left skimped out and took the easy route, whoever made the one on the right didn't finish it.
>>
>>637356
the key is pulling all normals in one direction so your shading constant
>>
>>637356
The render on the left looks too fake. The highlights in the hair seem to be part of texture and I think that because of the shading it appears to fade out. It looks inconsistent compared to the right. Basically the left has stock/viewport shading and the right doesn't.
Also that flower. What the fuck.
>>
>>637356
It seems on the right the shading around the mouth area breaks
>>
>>637369
>>637371
>>637372
>>637373
Thanks, good responses.
>>
Mandatory Guilty Gear mention:
https://www.youtube.com/watch?v=yhGjCzxJV3E&t=14m20s
https://www.4gamer.net/games/216/G021678/20140703095/

Is this still the gold standard?
>>
>>637384
https://www.youtube.com/watch?v=UYEsJLiqBw8 could it be they have finally figured out how to do it in 3d fighting games too?
>>
>>637384
would this work for someone using motioncap though?
>>
>>637400
l don't see why not.
Mocap is fucking garbage tho.
>>
>>637356
textures and shaders
>>
>>637400
Not automatically and completely. The stuff about texturing the base model and stuff would probably be ok even in realtime, but the most expressive poses and animation are achieved by tweaking certain frames with a particular camera angle in mind. You could edit mocap footage like that I guess, but if you're going for anime aesthetics you might as well animate it like anime (traditionally).
>>
>>637356
The one on the right looks like it's being lit by the scene and casting a soft shadow around the mouth, chin, and cheek.
That clashes with the hand painted shadowing in other areas, and clashes with the toon style (specifically the anime style; you can have a non-realistic or stylized render and have your own shader, but it's all about consistency).
The eyes don't look that great either, they look really flat and the painted on eye reflection looks bad and doesn't match the light source on the hair (the intensity specifically I think, not the direction).

Left doesn't look that great either, looks better than right, but left has some weird stuff going on where there's a lack of detail due to style but then you've got the coloration on the cheeks trying to create definition and shape but it's sort of wrong because it's not hand painted (or it's hand painted but it's just a simple brush).

There's some other stuff going on as well, all of it really boils down to many many little details that either clash, are inconsistent, or done poorly; if you get enough such little micro errors the render becomes unappealing even though it becomes very hard to say why because the overall form may be "good" (as in objectively technically proficient and adhering to an intended design).
>>
>>637500
>right looks...
>left doesn't look that great either
This. I had no clue which one OP thought looked good.
>>
File: AOI.png (269 KB, 464x431)
269 KB
269 KB PNG
>>637517
>>637500
it's strange because in the japanese youtuber fandom Aoi is considered to have one of the best models. i get what you mean by unlit / no detail, but people like it and consider her "perfect / pretty / 2D" looking.
>>
>>637387
Everything from the Guilty Gear pipeline, with the exception of the perspective deformations, applies to full 3d POVs just as well, so there pretty much was nothing to figure out (and you can still apply the perspective deformations whenever you know where the camera's gonna be).
>>
>>637400
We've known since Walt Disney and the early rotoscoping days that merely replicating human movement isn't enough. Any big game that uses mocap has artists touching up animations after that, so it changes nothing.
>>
>>637579
>We've known since Walt Disney and the early rotoscoping days that merely replicating human movement isn't enough.
Can you elaborate with a few examples, or point to an article or book explaining why? I'm genuinely interested in what the shortcomings may be -- unless we are talking about cartoony animation, where it would be obvious.
>>
>>637356
right looks more 3d and it's something that you don't want there
>>
>>637582
Start by reading the illusion of life, for example. The basic idea of the shortcoming is recording human movement is not as precise and not as complete as we'd like and subject to the uncanny valley, so extra work has to be put in to make it larger than life and offset that effect. Was true for rotoscoping, is still true for motion capture.
>>
Have there been any attempts at (more or less) realtime mocap stylization? I guess the biggest issue is probably a lack of high quality data. Then again, there's probably a lot of generic processing that you could apply, like suppressing minor movement.
>>
>>637579
this.
When did we forget our love for the 14 principles of animation? Mocap may be more precise and accurate, sure; but is that what is most important.

I'll take an animator who was able to act through his figure using only his knowledge of balance and motion, rather than just hiring kiefer suverland to crap out some stale actions and shit. Your craft is headed to hell on a highway and you guys are too busy trying to make a weeb shit that should only exist in 2 dimensions.
>>
>>637637
I thought Kizuna Ai was basically that, some version of this kit. You have to roll your own model though.

https://neuronmocap.com/content/product/32-neuron-edition-v2
https://kila.amebaownd.com/
https://twitter.com/livecartoonjp
>>
>>637637
Why stylize mocap data when you could just hire traditional 2D animators to do their usual job but replace inbetweeners with pose matching in 3D?
I think that's what they did in the japan expo short Rapid Rouge and it looks pretty good compared to most attempts at imitating hand drawn anime.
>>
>>637637
Suppressing minor movement has been attempted, but the problem is a lot of what makes a movement humanlike is precisely the minor stuff, so any such cleanup phase contributes to making the mocapped movement less convincing. Without a way to separate minor movement resulting from sensor imprecision from minor movement that should be there, it's only a hindrance. And resolving that problem is the holy grail of sensor science, not just mocap.
>>
>>637637
There's definitely research out there for stylization of mocap
http://research.cs.wisc.edu/graphics/Temp2/liyin/StylizedMocap.pdf
and I imagine big studios do some automatic stylization in addition to manual cleanup.
>>637665
Realtime-wise, even systems like perception neuron are basically just a constellation of IMUs, so there's a bunch of software post-processing that happens. You can consider that "stylization" to a degree, and you could use neural nets and training data to make that post-processing trend, e.g. more feminine/anime grill output if you wanted. Or more masculine, burly. There's a lot of stylization that can happen in the armature and IK itself for what you're animating.
>>
>>637356
Shitty lighting on the right.
The one on the left has a 2D looking face, meanwhile on the right you just think "yep, a 3D model".
>>
File: values.jpg (11 KB, 346x146)
11 KB
11 KB JPG
>>637356
The left one is actually more detailed, but whomever made the one on the right understood basic rules of color values for frame composition.
>>
>>637356
There's shading at all. You gotta mangle the normals on her face to get truly flat shading.
>>
>>637964
And yet it looks lame in comparison





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.