Home Technical Talk

Face Topology [Breakdown Guide]

2

Replies

  • SpeCter
    Offline / Send Message
    SpeCter polycounter lvl 14
    You should try to make the polydistribution more uniform, thats better for sculpting.
  • DrunkShaman
    Offline / Send Message
    DrunkShaman polycounter lvl 14
    I was hoping for someone to pick a red marker and correct the mistake I am making (if any for the blocked mesh's topology for sculpting).

    Poly distribution is visible after you soften the mesh to see how the topology flows. I have an updated version as well, I'll see if I can post it asap to help better understand my problem. =\

    femalefacecloseup.jpg
  • gray
    a few thoughts about topology.

    I tend to agree with some of the other posts that Dreamer's mesh has some problems. the mouth area layout is going to give you some trouble. you really do need some solid loops around the mouth with some extra loops bunched up in the cornes that can stretch out for a big open shape. I think crazyfool #53 mesh look about right for a generic mid res cage. notice there is a nice crease defining the nasolabial fold. but I have seen enough variation to have come to the conclusion that there is no real best way. you can get away with a number of configurations. if you can skin it, animate it and everyone is happy then its a layout that 'works'.

    you have to take those Pixar meshes with a grain of salt. VFX studios and toon studios have two completely different philosophies about models at this point, and different needs. hi-res games and vfx seem to have about the same workflow. sculpt everything in zb/mb resurf then extract all your detail in maps. the only difference being film uses displacement. and games normal maps. toon on the other hand really has no high resolution local detail. the detail is all texture detail. you will actually find people still 'box' model in polygons for toon characters. if your end result is a simple smooth character sculpting is for the most part extra work. and also if your not going to pull maps there really is not much point in sculpting. what they do have is a HUGE amount of radical deformation. crazy shape that you can't really do with shape targets. you need live deformation systems so you can make shapes on the fly. and put some of that ability into the hands of the animators so the can custom tweak shapes for there shot. so most of the modelers time is spent reworking edge flow to give the rigging team the best mesh for there deformation needs. and probably more then one version of the same character.

    also what are those pics of, a wip? final mesh? maybe a test mesh that the hair team was using to dev hair. a driver mesh that controls deformation for a final render mesh that has a different layout? any given character will have many meshes associated with it for different tasks. there is a different mesh for animation. cloth sym, muscles, etc. all of this gets blended together. so if your interested in good topology for hires humans and creatures the last place to look is at toon models. and you have to actually know what type of mesh your looking at or you could totally miss judge things.

    probably one of the best tools is topological asymmetry in the mesh. this is hard to do on a budget. but you could probably shoot for deformation asymmetry with texture asymmetry to get some really solid looking believable character in the face without pushing poly count to high.

    @Nitewalkr
    I would try to shoot for crazyfools layout. but you should concentrate on the art of making better characters not so much on poly layout.
  • JacqueChoi
    Offline / Send Message
    JacqueChoi polycounter
    Almost everyone missed this (cept crazyfool)

    stock-photo-10103393-amazed-beautiful-young-blond-woman-with-open-mouth.jpg

    The laugh line that runs from the side of the nose past the corner of the mouth is supposed to loop UNDERNEATH the chin. Not undernearth the mouth.

    There's a few other instances of how I do 'Crows Feet' Angry lines, and overall wrinkle topology that follow skinfold principals.
  • kostyasazonov
    Del, thanks a lot for your tutorial. Really helpful images.
  • Anchang-Style
    Offline / Send Message
    Anchang-Style polycounter lvl 7
    I found this to be very usefull.
    http://www.phungdinhdung.org/Studies_paper/Realistic_face_modeling.shtm
    There is a lot about mistakes.
    But with most tutorial videos, especially the Seraphine one, is what to do with the end of the lips? where to connect them? thanks to the extrem low res of the videos you can't see what the artist is doing.
  • malcolm
    Offline / Send Message
    malcolm polycount sponsor
    How come we haven't included the Gears 3 face topology on the wiki?
    gears3_topology_01.jpg
    gears3_topology_02.jpg
    gears3_topology_03.jpg
  • franman
    Offline / Send Message
    franman polycounter lvl 13
    Hi I would like to share a face model with you. This is a work of mine about to be ready to be imported to Mudbox. Comments & critiques appreciated.
    malehead2.th.gifsidefo.th.jpgfrontxhz.th.jpgbackwtq.th.jpg



    angle4.jpg
    By franman007 at 2012-02-23

    angle1g.jpg
    By franman007 at 2012-02-23angle3.jpg
    By franman007 at 2012-02-23
  • Sergi Caballer
    I would like to share with the to the Polycount's user community a couple of timelapses which I've been working on for a personal project. I opened a thread related to this topic here

    I hope these videos could be useful for somebody! ;)

    [vv]63320058[/vv]

    [vv]64302836[/vv]
  • artquest
    Offline / Send Message
    artquest polycounter lvl 13
    Very nice Sergi! That character is quite awesome. I couldn't stop watching your time lapse. Thanks for putting it together :D I wanted to ask, what tool are you using to shape the eylids at around 1:43? The pink wireframe leads me to believe there's something besides soft selection and the translate tool going on there.
  • Sergi Caballer
    Thanks for your words artquest!
    Definitely you have put a lot of attention to my time-lapse! Nice! As you mention I was using a deformer at around 1:43, in this case was a lattice deformer to give to the eyelid the proper shape to adapt it to orbicular shape.
  • vargatom
    Hey guys, we've been interviewed by FXGuide recently and I've managed to sneak in two wireframe shots.
    http://www.fxguide.com/featured/how-digic-made-the-assassins-creed-iv-black-flag-trailer/

    First is the kind of topology we use for the hero characters:
    blackbeard_wireframe.jpg

    This is basically what I've been able to reverse engineer from Weta's approach on Avatar and the Hobbit and implement with our budgets and schedules in mind.
    It's relatively high res so that we can sculpt nice blendshapes. Also, the edge flow is based on the character's facial features and aligned to the major crease and wrinkle lines on the concept sculpt. In case of the neck the flow is aligned to the platsyma muscle and the crease lines under the chin.
    BB still isn't too old or ugly to have a more mixed topo, but on another character I'm working on, the sides of the face have more Y-patterns.
    I think BB is around 16000 quad faces and the mesh goes down to the clavicle bones.

    Then there's the generic mesh we re-use for the background characters:
    pirateAdewale_wireframe.jpg

    This one I wanted to keep as simple as possible but still allowing for blend shapes. No facial wrinkles or creases beyond the nasolabial though. We will do small revisions to the neck area because of the skinning, and I might add another loop to the brow area. We're also going to look into animated displacement maps for facial wrinkles later on.

    Maybe this can help some of you guys, particularly as you start with the nextgen projects...
  • katana
    Offline / Send Message
    katana polycounter lvl 14
    Great vids Sergi...Thanks.
  • artquest
    Offline / Send Message
    artquest polycounter lvl 13
    Thanks vargatom, those wireframes are truly awesome.

    I have a few questions for you! :D Where did you find the references from avatar to reverse engineer from? The ongoing discussion of facial topology for animation is very interesting to me! I love the setup of these newer topology :).

    What are some of the considerations that lead to this change? Is it purely used on higher polygon meshes intended for high fidelity animation or will the edge flow displayed on blackbeared work with lower polygon meshes(say game meshes). Are the benefits lost when the polycount is reduced?
  • vargatom
    Hey, you guys are welcome, I've gotten a lot out of online communities including polycount, so I just try to give back as much as I can ;)

    The Avatar stuff came from a Yahoo behind the scenes movie, I think this one:
    http://www.traileraddict.com/trailer/avatar/extended-featurette-creating-the-world-of-pandora

    There were a few images of Neytiri's model and also some footage of dialing in blendshapes for Jake's avatar; these got me thinking, and it also helped that I've read the Maya master classes of Jeff Unay who was the facial lead at Weta.
    You can also check out Weta's youtube channel for some BTS movies on Gollum, the Goblin King and Azog to see the same system in action.

    The main reason for this approach is to get all your facial deformations done in the mesh and have no animated displacement or normal maps. It is the right way to do it if you build a mesh dense enough, to simulate the way skin folds and wrinkles are formed.
    It's also much easier to create the blendshapes using just brush based sculpting tools, and the modeler has to spend less time with terminating edge loops if there's no need to keep the poly count down.

    The game changer is that nowadays we have computers fast enough to work with such complex models, but it wasn't possible 5 or 10 years ago.


    So how this can relate to games.

    Current gen solutions are usually based on bone based rigs and some way of blending between two or three sets of normal maps which reveal facial wrinkles. Since most games are also using a large cast of characters, many studios build one base mesh and set of UVs across everyone, to re-use as much data (like skin weights) as possible.
    This is also why I decided to re-use meshes, UVs and blendshapes and looked at games for reference because those guys have really perfected this stuff.

    Now, I've attended the talk on Halo 4's approach at FMX this year, by Matt Aldridge, which I believe is the middle step from current to next gen. I hope he doesn't mind me posting a brief summary here.
    They are re-using the same base mesh and UVs, but they use blendshape rigs instead of bones, and scan real humans to get the maps, the expressions and even the textures. Then they fit multiple Zbrush subdiv levels of the base mesh to the scans; the level 1 is the actual blendshape of the rig, and the level 3 or 4 mesh is used for wrinkle maps.
    (the way they got so much data into the game was that they did not import the animation rig - they used Principal Component Analysis to get 1:60 compression on the deformations)

    However, their capture process gives them enough data to scale the stuff up to next gen, and use the level 2 or even the level 3 base mesh in the game. Which would be either ~10K or ~50K quad poly meshes for the faces, although I'm not sure how that would shade without subdivision and/or a normal map on top. So they may not be able to skip the animated normal maps or not.
    Also, a generic base mesh that is not aligned to facial features could prove to be difficult to sculpt facial wrinkles and folds the way it's possible with the approach used by Weta. On the other hand they could use tessellation and displacement maps instead of normals to recreate the geometry from their scan data.


    So what I think applies to nextgen games based on the above:
    - 10 to 50K polygons without subdivision may not be enough to get nice enough shading on sculpted skin folds/wrinkles in a game engine
    - building a custom base mesh and UVs and blendshapes per each character may not fit into the schedules and budgets of a video game

    I'd also call attention to the Activision tech demo based on Lightstage scan data, using a bone rig and various wrinkle maps in both the color and normal layer. It is pretty damn good looking if you ask me and the base mesh seems to be a generic one.
    So, it doesn't look like the Weta approach can be beneficial to a realtime engine; and even if you're not relying on scanning you could manually sculpt source data and process it in a similar way. Which means the base mesh on the generic pirate is probably more relevant, but the resolution could be increased as current gen games are already at similar levels.

    Hope it makes some sense ;)
  • vargatom
    I've just seen this yesterday, character rigging in Last of Us:

    [ame="http://www.youtube.com/watch?v=myZcUvU8YWc"]Judd Simantov on character rigging and modeling in Naughty Dog's The Last of Us - YouTube[/ame]
  • Sergi Caballer
  • Ahoburg
    Hey guys, forgive me my nooby questions, but i just gotta ask. How does very detailed retopology of a face (sculpt or scan) works? Is it just so many polys in it because of subdivisions or people retopologize the face like this from the start?

    Also, can anybody show me the way to some decent face topology tut videos? Thanks.
  • oglu
    Offline / Send Message
    oglu polycount lvl 666
    this is a good example of the workflow...
    https://www.youtube.com/watch?v=WJFu26pWxVI

    you could do a lower version like the one in the vid, subdivide it once and reproject it... or create a denser version from the befinning... depends on the model and your workflow...
  • Sergi Caballer
    Hi guys,
    this week I uploaded the 3rd part of my facial modeling timelapses, and also a website where I describe the overall process followed. I hope you like it!

    http://www.sergicaballer.com/facial-modeling-timelapse/

    [vv]73340876[/vv]

    Cheers!
  • ajayp
    Offline / Send Message
    ajayp null
    Hi guys,
    the Images in this post doesn't load now.(I think the links broken)
    So please whoever the owner of this thread can you please re-link the images....
  • AtticusMars
    Offline / Send Message
    AtticusMars greentooth
    ajayp said:
    Hi guys,
    the Images in this post doesn't load now.(I think the links broken)
    So please whoever the owner of this thread can you please re-link the images....
    As I recall Del's original topology didn't have loops to support good deformation around the mouth. If you need a guide I would suggest using Crazyfool's image up above.
  • Eric Chadwick
    We have those examples and many others here in our wiki.
    http://wiki.polycount.com/wiki/FaceTopology
  • RN
    Offline / Send Message
    RN sublime tool
    There's some more examples from the Blender community. All these were done by the same artist, Angela Guenette:

    Protagonist from "Sintel" (2010)

    Blenderella course (2011)

    Victor from "Cosmos Laundromat" (2015)
  • vvvickyyy
    Can you update the images? This is a great resource, I provide it to my students. 
2
Sign In or Register to comment.