G'day mates,
The Overwatch beta was a blast and the game looks simply gorgeous.
I decided to grab a bunch of screenshots and a few ripped models and have a look at how Blizzard are making Overwatch look the way it does.
A lot of this is just speculation, I very well could have gotten something wrong in here.
To clarify, I did not work on Overwatch, that would be amazing though xD
I also compiled the slides into a PDF for those that want to download:
(EDIT: Updated link, contains some corrections)https://drive.google.com/file/d/0Byjlw-rmzZZwQTFvQkpobVpqUUk/view?usp=sharing Also posted it on my Tumblr if that's your jam:
http://agiledev.tumblr.com/ ]
Note: These slides are a little out of date, check the PDF slides for a slightly updated version with some of the corrections you guys have suggested.
Replies
Cheers
That unidentified D.Va hair texture is an anisotropic direction map to control how the highlights in the hair shader behave.
Love the comments on Intersection meshes, corners and VFX!
Hopefully once OW is out I can do a version two of this breakdown with some of the info cleared up so it's less speculative.
Cheers!
they did not erase the blue channel there but also seemed to change them somehow to this green look in a more complex way
This is done to make it possible to create easy overlays of normal maps, so you can blend different normal maps into one texture without
losing much detail (like for additional skins maybe?) so they dont have to make all new normal maps but just blend a new one over
the old.
In the UMA2 this method is used to add scars, hairstyles, etc... to the meshes without needing several different combined normal maps.
So they just combine 2-3normals and overlay them still having a clean look and not washing out any channel
this could also be something they were thinking of?
Thanks!
For example, the tertiary map likely uses BC1. The metal mask is pretty broad and constant in large patches, so it can get by with the lower precision R5 channel. The specular, on the other hand, clearly has a lot of smooth gradients since it has specular occlusion information pre-baked into it. So they stick it in the higher precision G6 channel, and then mirror it in the B5 channel to reduce artifacts from the three channels being compressed together.
The normal map is only using RG since it will likely go into a BC5 texture. The heightmap is only using R because it will likely go into a BC4 texture. The albedo map is likely using BC3, since it has an alpha channel for Ambient Occlusion. Finally the emission uses a BC1 since it is just color data.