How do Animation Studios texture?
interpolator
I've been trying to find something about how Disney or Pixar or the likes texture all their environments or characters but I can't really find anything.
Or even environment breakdowns ,all I get is animation breakdowns.
I was curious if anyone had an insight of how they do it. I have a small inkling that it's maybe only shaders but I'm not a 100% on that. ( Or if they UV unwrap as well )
Or even environment breakdowns ,all I get is animation breakdowns.
I was curious if anyone had an insight of how they do it. I have a small inkling that it's maybe only shaders but I'm not a 100% on that. ( Or if they UV unwrap as well )
Replies
http://youtu.be/NT_FYRP7kqI
Apparently disney also wrote a paper on that so I might have a read through there as well ^^
How do you mean?
Pixar use Mari with UDIM and PTEX layouts.
Dreamworks and Sony Imageworks use Mari with UDIM layouts, ( as do Weta, ILM, Framestore, DNeg, MPC, Cinesite, Blizzard Cinematics, Blur, MPC)
Their UDIM layouts can run into many hundreds of patches per character or asset, normally using a 2k or 4k texture per patch.
I could probably answer any questions people have about high end VFX texturing.
When I see the credits I don't think i've seen the titles Texture Artist in Frozen/Tangled/ etc.
1. Is texturing for stuff like cartoons, popular in VFX? What does the workflow look like?
2. What is it like texturing for something photoreal? Are there any special rules, do you let the technology do most of the work, and what about an artist own input?
3. Since high end VFX is years ahead of video games, what kind of technology can we expect to fall into game artist hands over the next couple of years?
With PTEX there is no layout process as each face gets it's own texture. It's nice to not have to UV anything.
It depends on the company as to who does the UDIM / UV layout. Some companies have their texture artists make their own, and in others it's a modelling task.
Tools like Headus UV-Layout are commonly used to speed up the process. Pretty much essential when your character has 650 patches.
Most of the time modelling and texturing are different jobs. It's not usual for the modeller to do the texturing too.
Look development is the process of linking textures to shaders and balancing those shaders to achieve a final "look". In some companies one person paints the textures "Texture Painter" and another does the Look Dev "Look Dev TD". Other companies will join those roles into a single job (Sometimes called Surfacing or just Look Dev).
Disney joins the two tasks and credits the people painting textures as "Look Development Artist"
Dreamworks does the same and credits people as "Surfacers"
Weta credits people as "Texture Painters"
1. I'm not sure I fully understand the question but I'll write some stuff anyway
Companies like Pixar and Dreamworks will paint textures to achieve a cartoony look. It used to be much more common to use procedural shaders that are evaluated at render time rather than textures. These saved on memory and disk storage, but came at added complexity for things like anti-aliasing and flexibility. Although procedurals are still used, hand-painting is much more common than it was.
The normal process is that a concept artist will prepare a large number of look studies, including things like colouring, patterns and surface detail. They may also provide a large number of reference photos of skin textures that they want the texture artist to match. This forms a "look pack" in Pixar terminology that is then handed to the texture artist.
The texture artist will then match the look by painting diffuse, spec, bump, etc etc etc channels. This will be a combination of handed painted and heavily modified photo reference.
2. Photoreal texturing takes two forms. Prop / environment matching and entirely synthetic textures.
This might get a bit technical. Sorry.
When prop matching the texture artist will ideally be provided with a large amount of professionally shot reference material. I.e they took the prop into a studio and shot lots of different images from different directions under controlled lighting conditions. If the reference photographer is doing a good job they will include a macbeth colour chart in the shoot.
A Macbeth chart is a piece of card printed with different squares of known colour values. Having this chart means that you can balance the photos, i.e. adjust for any colour changes introduced by the camera, lens, sensor or lighting. If you know what colour a pixel "should be" you can apply colour transformations to ensure that it is.
Ideally the photo shoot will also use crossed polarised filters to remove specular highlights. This is a technical way of shooting an image with two filters, one on the lens and the other on the lights. By using these filters you can recover only the diffuse component of the colour which is what you want if you're going to texture with it. You don't want any specular response in your textures as this will be added later by the shaders at render-time.
Becomes
When cross polarised.
Once you've got your balanced, diffuse-only reference, you would then use a package like Mari to 3D project that reference onto your mesh of the same prop. If everything works out the model is accurate and you can align your reference shots exactly. Then it's a matter of creating plausible spec, rough gloss etc maps to recreate the surface response of the given prop.
When you're working without a physical prop, but with an entirely virtual object, you'd normally source reference images that match the materials you're looking to recreate. I.e. high res metal, steel, concrete images. You'd then use Mari again, with lots of warping, stretching and other adjustments to "fit" the reference to the surface.
Doing all of this in 3D means that you don't need to worry about seams.
3. 3D painting will become much more common. Dealing with seams is a paint in the ass and there is no reason to put up with it. Most modern CG movies are almost entirely textured in 3D in Mari with PS being used only when it absolutely needed.
Painting in the context of a live shader will become more important. As physically based shading becomes more common and shaders become more complex it will become more and more difficult to efficiently create a look using a normal 2D workflow. When you're painting esoteric maps like subsurface scattering, different gloss response controls etc. it's very difficult to judge what effect a mask will actually have when you run it through the shader. Programs like Mari allow you to run your own shader while painting, so it's a much more WYSIWYG experience.
Beyond that, bigger textures, higher bit-depths and a greater understanding of colour management techniques and colour space handling.
Outside of Disney PTEX is most commonly used for environments. When dealing with complex, especially, scanned geometries it's a god-send not to have to UV them. They're sometimes a mess.
Minimising texture sizes is no-where near as important as it is in games. Disks are cheap, memory is cheap, network is cheap. People are expensive. If working quicker means you are heavy handed and don't really optimize it's not really *that* much of a problem. It's better to over-paint and never need to add more detail than to go back and spot update things.
A large environment like Rivendell from the Hobbit might have upwards of 30,000 8k textures. It isn't just environments that go a bit crazy....
I was talking to the lead artist on a very well know creature from a recent movie and she told me the creature had 650 UDIM patches and 30 channels (diffuse, spec, bump etc) each at 8k x 8k which works out to 19,500 8k textures or nearly 300,000 2k textures.
Most of the time everything is uniquely textured. It's quicker to create and less complex to manage. The texture setup is "put these textures onto that object using this shader" not "tile this x times and this y times, with this blending mode etc. etc."
Do they bake stuff too? Or do they directly use model with the highest polycount?
(Also a personal question if I may. Where do you know all these things from? Work, websites, school?)
jgreasley: That sounds like a crazy amount of data. Do you steam it in as needed or just rely on crazy spec'd out systems... Or rely on network rendering or something?
These tiled textures allow to load only the required mip levels of each UDIM texture.
Here´s a quote from the solidangle documentation
Another point is that we don´t store the real models in our scenes so they become really light. We use Referenced StandIn Files. These are basically precompiled rendercode Files created for each static asset. For Characters we create Cach Files which are loaded onto the Render Geometry. So The render Scene has no Rigs, Skins, Controls, or real Meshes.
Yep, Mari will render all of those textures in realtime. You need a bunch of fast storage / SSDs / FusionIO cards to work efficiently with that much data, but as long as your disks are quick enough, Mari can handle it.
In general geometry is represented as Subdivision surfaces which then have sculpted detailed applied on top.
The base mesh will be modelled in Maya etc, and the detail sculpted in Mudbox / zBrush or painted as maps in Mari.
Ideally the model will be represented as a relatively low res subd cage mesh and a collection of one or more displacement maps. The renderer will subdivide and displace the model at render time. This means you don't have multi-million poly models flying around. You have hundreds of thousands of ploys and displacement maps.
Some facilities will split displacement into different frequency bands (coarse detail, mid detail, fine detail) which allows different features to be adjusted individually. An example might be having skin pores and details on their own maps, so you can balance them in close-ups etc.
As to knowing this stuff, I've worked in VFX for the last 13 years and I wrote Mari.
Yep, Exactly what .Wiki said. All modern production renderers have tiled caching schemes. They will page in just the tiles that it needs as it needs them. You need to pre-process the data into tiled, mipmapped versions, but doing that makes it possible to render these huge texture sets in realtime.
A close analogy would be ID Tech 5 that was used in Rage. If you watch some of the videos about the Megatexture technology this is pretty close to what Renderman et al do to manage textures.
Great, that makes a lot of sense, thanks! (Also Wiki, thanks!)