Arslan: Warriors of Legend

In an attempt to better understand how Omega Force at Koei have managed to make such a good looking anime game, I decided to also get my hands on a PC version of it to rip it apart. This is just the first stage of ripping, and further posts on this game will be made soon.

This image already gives us a good idea about how the game is textured to give the anime effect. The textures are very simple and largely block colours on the characters, while the environment looks to be either heavily filtered photos or hand-painted textures, not to dissimilar to other titles from Koei.
The image shown above is actually not the right colours as seen in-game, which is strange. XnView does however, see the image file as it appeared originally. The file does have an extra channel showing a cut-out mask for a cartoon-style, but applying this with any kind of layer styling still doesn't recreate the way it looks in game, so I'm not sure how it works yet. It looks as though the layer is simply set to Darken or Multiply, but that doesn't correct the colour, nor is the colour a direct invert of what is seen above.

This is the shot from XnView:

Here we have the specularity pass. Notice again that the environment is really somewhat standard, while the characters themselves appear to not use a dedicated specular map. Instead, they are simply filled with white, which would give a very flat, shiny look if they weren't cell-shaded afterwards.

We can see however, that everything in the game world still uses normal maps to give a fake sense of resolution. The characters of course also use these, but the results of the bump are posterized to fall in-line with the anime look.

There is also a separate ambient occlusion pass to the game renderer, but note that this is not it. This came out along with the diffuse colour pass as a 4th channel. It appears to be some kind of mask that is used to cut the shadows and highlights up to make them look like an anime, which is interesting. 

------------------------------------------------------------------------------------------------------

Initial finding from this first rip show no signs of actual texture maps that aren't related to particles, which is odd. The textures shown on the characters appear not to have been ripped, nor do the models. In fact, the only models that came out were particles once again, which is odd.

Tomb Raider: Underworld

All work shown below is subject to Copyright © Crystal Dynamics 2008. These models are from the xBox 360 version of the game data.

Alister

Characters from 'Tomb Raider: Underworld' were created with many texture maps, much like my other current generation game research shows from other games such as 'Tekken 6'. This certainly seems more common practice now than using atlas texture maps. Maybe loading lots of smaller textures works out to be more efficient now?

His mesh varies in detail levels, mainly focusing on folds and bends that would need to animate: something that is true for all models. You can see how his mouth sits inside his head, and how his eyes are constructed. The eyes are actually made up from 3 layers of polygons, not the more commonly seen 2. Instead of having the eyes themselves, then a glaze of shadows and highlights on top, there is a 3rd layer, but it is unclear as to what is on this layer. His texture files suggest nothing and all that I can think of is that the 3rd layer must just use a special shader in-game that can be used to lightly reflect a set image (fake reflection). If this is true, why is this process not applied instead to his shadow / highlight polygon panel over his eyes? Or would this cause their effects to become distorted also?

Amanda

Amanda has some great fine details with her normal maps, especially around her knees and elbows where he coat goes under her gloves.

Notice the level of detail seen on her textures. Like many other characters seen in todays games, she fakes the look of much higher resolution textures by using a 2nd set of normal maps that are set to repeat all over the selected mesh on the character. In this case you can see her detail normal map on her trousers, and a 2nd one on her coat. It is a brilliant technique, but currently is not controlled very well in most games (none that I have found yet). For example, using something similar on skin often causes the 'skin-pore' texture to apply itself to the lips too, making them look very strange close up. There are ways to overcome this, such as simply putting the lips on a separate polygon mesh over her face. However I wonder if for larger issues and even more control, whether making a 'masking' texture to say where the detail map should and should not apply itself would be better. Of course, this kind of thing would require yet more texture memory, and make the shader with it applied too much more complex to process. This arguably starts to fall into the realms of why alpha textures are kept at a minimum as much as possible.

Natla

A character that makes very obvious use of back-face culling, with only a single polyplane running through her wings to represent the skin between the joints. Dependant on the angle of the camera, depends on whether or not one side shows, or the other. It is a very common and important technique used in 3D to represent things that should not really have any noticeable width, such as hair planes, capes, and other hanging fabrics for example.

Notice her hand and the way her fingers are positioned. They have as little bending as possible, and the thumb is also kept fairly straight. This process allows for the hand to be more easily rigged with more complete movement afterwards. It could be said that modelling the hand slightly closed would give better end results for a character who would always have his hand closed or a very similar position, but if the character needs the full movement motions then a standard pose such as this is more required.

Winston

Winston again makes use of the detail normal maps to a larger extent to detail the skin on his face. Also notice how the bottom of his suit jacket flares out oddly. This is down to how the physics engine in the game would react with it. Spreading it out like this not only makes is easier to rig, but it also gives it a good distance between its collidable object (the character), and itself. This ensure in-game that it falls down into a natural position well, whereas modelling it in a relaxed position to begin with could produce abnormal and even adverse results.

Winston and Alister both had their hands in this open-fingered position. While there is nothing wrong with this so much, it does make you wonder why the women have their fingers close together instead. I believe this is down to the different rigs. For natural movement, it would make sense that the males and females both have separate rigs. These rigs were likely built by different people during the games development, and thus you get different layouts. The modellers would have to adhere to these layouts to make sure their characters fit perfectly to them.

Dynasty Warriors 4: Empires Research

Rendered in Maya's Mental Ray, these characters were ripped from the archive section of 'Dynasty Warriors 4: Empires', posing the characters, freeze-framing them, then ripping the data. Although the Playstation 2 does some perspective changes in real-time, causing the models to be slightly skewed in places, they still show the topology and textures; valuable information for how they were built.

The first face here of 'Lu Bu' shows this distortion well, but the geometry is still very much readable. You can see how the faces of the characters were still made to be low-poly, but still contained some rudimentary geometry to animate reasonably well. They do not follow current generation edge-flow techniques, but still maintain a high quality appearance as the renders show.

The textures in the game only include Diffuse maps because the Playstation 2 was incapable of handling other maps like Normals, Specular, Ambient, etc. However these maps were often not square. The above image shows two maps, separated by a black line towards the right. You will notice the texture on the right is very tall and thin, at a 1:3 ratio. The reason for this is the ability to be able to get more onto one map without the need for extra textures to be loaded, whereas nowadays we often see separate small textures being loaded for each part of the body, such as the head, hands, shoes, etc. The very tall texture on the right is loaded just once per level, containing all weapon textures in the game: a much faster process than having to load a new texture every character with a different weapon. The downside to loading textures with aspect ratios that are not equal is that they take marginally longer to process, but you can imagine that it would still be faster for the console to load than having to load multiple square textures. This kind of optimization is still in use today, but we don't see it as often for obvious reasons.

Case-Study: 'Kazuya Mishima' from 'Tekken 6' - Character Texture Handling

This paper was created to study the model of 'Kazuya Mishima' from 'Tekken 6'. You may download and view the paper for free, and use it for purely educational purposes. The link is below. Click to view, or right click and save to save the PDF to your computer.

Next - Current Generation Game Character Texture Handling (April 2013) - Ricky Thomas

Tekken 6

Tekken 6
Playstation 3 / XBox 360

• Model Data

Many current generation games are catching up with eachother in terms of polycounts for characters. 1vs1 fighting games no longer hold higher-polycount characters, simply because they don't have too. All games (within reason) often have character made up of 12k - 20k tris now, no matter the genre. For example, 'Uncharted 2' had between 15k - 27k polys per character at their highest LOD. Tekken 6 is the same, and a great amount of detail is conveyed through this. It is getting to the point where higher resolutions are almost unnoticeable, and thus, a waste of time and memory to make them.

'Asuka Kazama'
14,779 Tris

• Extremely high detail modelled into her face in particular, allowing for great animation. Edge-flow wise, her entire body is very good for deformations. The bandana around her head is modelled straight out of the back on her head: the physics engine in-game will make it fall naturally. This is something that we see all of the time nowadays.

• As Namco often or always did even back in the PS2 era, the character has around 10-15 textures. The model is split up into many different bits that all get their own texture space, usually of a size around 128-512, depending on importance and model size. This method also appears to be quite common practice in most games now.

















'Steve'

18,360 Tris

• His model is very well detailed, with edge-flows following all of his major muscles. While this is clear, it is hard to tell if this was originally all quads, or if it was triangles that made up such details in the first place.

• Having every major muscle outlines with polygons does make sense for flexing each muscle, but this is not the case in-game. In a way, it almost seems wasteful because normal maps could convey the information for less memory cost (not that I'd imagine the console would be struggling with memory in such a game).

• Again, just like 'Asuka' above, he has several textures. One thing I notice is that the specular maps *appear* to be red. Maybe this simulates a fake sub-surface scattering effect? I'm not sure, but soon I will try to apply all of the textures to one of these models to see what the outcome is: that should answer that question.

• His boxer shorts have a problem at the bottom whereby there are two polygon groups overlapping. I'm guessing this is how they make the models still look right with backface-culling on, whereas other games often just turn it off on characters. This way is technically more efficient though.













Overall, the models are extremely detailed. I have several more characters from the game, but these are just examples of two of my favorite ones. They are interesting to look at up-close: you notice all kinds of weird practices that suddenly start making sense. In general, it seems characters in most games will stay this high resolution for some time now. I don't think more polygons are needed on current character, just better texture resolutions now. A possible future for characters I guess would be real-time deformers for muscles, wrinkles, etc, but for now, normal maps work for this just fine.

Soul Calibur II

Soul Calibur II
PlayStation 2

• Model Data

Characters in Soul Calibur II are roughly twice that of other 'standard' games released on the PlayStation 2. By 'standard', I mean games that typically have several characters on screen at once, along with complex environments and AI; 3rd-person action-adventure games are generally a good example of this concept.

This game does not use backface-culling. The extent of polygon optimisation visible in the game is that it doesn't draw things that are not on screen. For example, the surrounding environment has polygons clipped as they go too far off-screen. However, notice that not even the environment uses backface-culling.

'Ivy'
8711 Tris
(Roughly 4355 Polygon Quads)

• Her model follows current-generation standards quite closely, allowing for some, at the time, advanced animation to be played on her. Examples of this would have included facial animation (she has the inside of her mouth modelled in reasonably low detail).

• Her weapon comes apart as several pieces to act like a whip. Generally items such as weapons are separate from the character model files, and have their own set of rules; the characters simply play the correct animations according to the weapon state (of vise versa; unknown which in this case).

'Kilik'
8635 Tris
(Roughly 4317 Polygon Quads)

• Again, he follows many rules topologically that current generation games do for advanced animation. However, one odd thing about his model that you don't tend to see very often is that his hair is modelled as seen, using no alpha maps, unlike 'Ivy' above.

'Taki'
8454 Tris
(Roughly 4227 Polygon Quads)

• Uses current generation modelling techniques, although personally I think she could stand to loose some polygons in specific areas of her torso, feet, and face. In game I don't think you would ever notice the difference if she did.






























Also I find this game to be very unoptimised, but having said that it does run absolutely fine on the PS2, so it doesn't need to be more optimised. The environments as well as some of the characters seem unnecessarily high-poly, and the textures are applied in a very odd fashion. Some of the characters have about 10 normal, diffuse (seeing as that's all the PS2 supported) maps, applied randomly to small parts of them at a time. Because of this, it would take a lot of work to re-texture the characters fully for this. The odd thing about this is again, linked to optimisation. Having the game load up hundreds of small textures instead of just the normal 1 or 2 for characters, would take far longer due to read/write calls. But again, the game seems to run fine either way.


In hindsight, I suppose games like this showed just how powerful the PS2 really could be at times. The real culprit as to why the graphics never looked 'amazing' was simply down to the texture types the PS2 could handle. For example, normal maps and specular maps would have made a massive difference to the graphics, but possibly equally to performance.

This screenshot shows the scene as captured from the game emulation, displaying just how unnecessarily  high-res some of the backgrounds were and yet the game ran just fine. I find this so interesting because other games released even in the same year as this were often incredibly optimised as if they needed to be, using every trick in the book to run faster. I wonder if Soul Calibur II could have loaded much faster if it was optimised as well as some other games I've seen, not that it seems to take particularly long loading as it is....

This image below shows 'Kilik' up close with his triangulated wireframe. However, you can still make out how he was modelled using quads for the most-part, and it clearly shows the edge-flow following around both his orbicularis oris and his orbicularis oculis making for good facial animation. Even the initial 5-point vert lies on his cheek-bone just as expected to see in today's games, joining his eyes to his mouth in an area that doesn't deform much with any facial expression. Despite this, the in-game facial animation wasn't... brilliant, but it did clearly make use of these modelling techniques to work at all. More polygons or animated state-controlled normal maps would usually help out with this sort of effect nowadays. In contrast to all of this, his ear is still typical of that seen on PS2 games, whereas today we would often fully model the ear also.

Below are two more renders of 'Ivy' and 'Kilik' standing together. The first is simply a beauty render of them, and the one below shows their wireframes, while also showing a problem seen on specific polygons on 'Ivy'. When the 3D data is given back as a .obj file, it requires a lot of clean-up to even closely resemble what it actually is. During this process, I always find that the normals are often messed up quite servilely. However, due to the fact that it is essential for the normals to all be facing the right ways in games (generally game engines only draw one side of every polygon (unless told otherwise) to save on render times), I can only assume this is a technical problem caused by the capture software being used to grab the 3D data. Nevertheless, I still must go in and clean up this problem. Normally, there are automated tools in Maya that allow this process to be fairly easy. However, for some reason this process out-right didn't work on 'Ivy' properly. The only real remedy for this problem seems to be to go in and manually flip the incorrect normals, which can take quite a while. Without doing this, the Mental Ray render engine has some problems when it is told to render the wireframe, as shown in the 2nd image down; the incorrect normals are unfortunately causing distorted pixelation on the appropriate faces. 'Kilik' never had this problem.

Welcome!

Welcome to the 'Research of Game Art' blog! This blog is here purely for learning purposes, both of its author and of the viewers. Posted here will be screenshots and technical analysis of game data ripped from video games, both from the past and present. The results will be presented for helpful knowledge when it comes to creating game art yourselves, including data such as model polycounts, texture sizes and specifications, and other such information. It is possible at times that some posts on this blog will be quite technical, and as such a prior general knowledge of how digital 3D creation software is used, is assumed (such as Autodesk Maya, Autodesk 3Ds Max, Autodesk Mudbox, Pixologic ZBrush, and so on). Along with this, it might be helpful to know some basics of different types of texture maps, and general video game creation methods. However, often things will be explained as and when they are mentioned. If you have any questions at any point, feel free to ask via comments on appropriate posts and I will get back to you as soon as possible.

Requesting Game Art

If there is a game currently on PlayStation 1, PlayStation 2, or PC that you would like me to analyse for your own research, feel free to request it and I will get as much information as possible when I have free time.