Human Topology for Rigging [Archive]

El Burritoh

24-04-2011, 11:57 AM

I'm usually modeling hard-surface and mechanical stuff, but I'm really trying to get into organic more and more, and specifically characters.

Here's a guy I've been modeling and I'm hoping for some feedback on the topology. Obviously the topological requirements depend heavily on what you need the mesh to actually do when it's rigged. Clearly, the following mesh isn't going to allow much detail in the areas of facial animation or muscle deformation. But my goal here is to have a good base mesh that can serve two purposes:

1) Accommodate a good rig so that the mesh doesn't deform unnaturally during animation.

2) give a better start during the sculpting process on human characters, and hopefully eliminate the need to retopologize certain characters.

I also wouldn't mind getting a discussion going about character topology for different kinds of rigs or animation techniques, especially in the areas of muscle deformation and facial animation. I know my example below won't allow much of that, but I'd love to see how other people are doing things.

Here is what I have so far. The reference for the face was "Ambrose" from 3d.sk. The ears are crappy right now...I'm still working on them. Hands and ears...I hate 'em! Oh, and if you have any comments on anatomy, I'd love to hear them!

Click on each for the high-res...

http://timcrowson.smugmug.com/photos/1262727074_zFV6cfL-M.jpg (http://timcrowson.smugmug.com/photos/1262727074_zFV6cfL-O.jpg)
http://timcrowson.smugmug.com/photos/1262727094_Xb6nxm7-L.jpg (http://timcrowson.smugmug.com/photos/1262727094_Xb6nxm7-O.jpg)
http://timcrowson.smugmug.com/photos/1262727124_3Hm7Xm2-M.jpg (http://timcrowson.smugmug.com/photos/1262727124_3Hm7Xm2-O.jpg)
http://timcrowson.smugmug.com/photos/1262727120_KmHNWqG-M.jpg (http://timcrowson.smugmug.com/photos/1262727120_KmHNWqG-O.jpg)
http://timcrowson.smugmug.com/photos/1262727108_vdXtDKv-M.jpg (http://timcrowson.smugmug.com/photos/1262727108_vdXtDKv-O.jpg)
http://timcrowson.smugmug.com/photos/1262727087_f3Q7Gs7-M.jpg (http://timcrowson.smugmug.com/photos/1262727087_f3Q7Gs7-O.jpg)

El Burritoh

28-04-2011, 08:26 AM

edited and bumped...

Kel Solaar

28-04-2011, 07:47 PM

Hi Tim,

I think that the less singularities, the easier it will be for the skinning and correctives blendshapes after, basically if you can mentally convert your mesh to big nurbs patches, I think the guys at rigging will be happy and your friends :rofl:

Actually your topology is quite close to the one I used on my Tendo:
http://thomasmansencal.com/Sharing/Pictures/Akatsuki/Breakdowns/Tendo/Modeling_004.jpg
It's also used to do the girl in my reel: http://www.thomasmansencal.com/resources/portfolio/pictures/marika/marikaFront.jpg
This topology while not being perfect has been done with my rigging friend and supervisor advices (Olivier Georges) for the feature film we worked on together last year.

It shares the same concepts than yours I think except maybe in some areas, especially the elbow. When I started to do it, I had the same construction than you, the arm being extruded to the bottom with the deltoid and shoulder being on another flow. Olivier made me reconstruct it to be like a sock, so that it handles better the deformations. I think the result to be quite effective.

The hips also displays the pelvis bone structure, it's not that important because most of the time, your character will be wearing something :D though it will make the skinning job quite easier because when the character lift its leg on the side, you won't have to do much diagonal weighting on the junction areas with the belly bottom.

Sadly I don't have the back, but if you imagine the neck being a sock again, the back is flowing like a tube from it, no singularities, nothing annoying for the weighting.

Voila :) Hope it helps :]

Thomas

BrettSinclair

28-04-2011, 08:22 PM

I cant say I've been in the field. But I have seen some vids. Jeff Unay did a master class with a wolfman type character where he goes into edge flow for muscle groups. I'd really recommend that.
What I can say though is it probably depends on what the character needs to do. Toon characters usually have a more simple edgeflow because they need to bend and stretch. Realistic characters need muscle bulges so you may need a more focused topology to achieve that result. I COULD be very wrong though :P

ulf

28-04-2011, 11:02 PM

yes, i think evenly distributed quadmesh, with less as possible singularities is best.
then it comes to what rig you want to do, what corrective morphs you need/Want at what level of detail. for that you may need to define more subshapes in your poly mesh.

El Burritoh

29-04-2011, 02:24 AM

Thanks guys! I'll try and find that material by Jeff Unay.

Thomas, yes, now that I look at your woman example, I see painfully clearly what I failed to do in the pelvis. I can fix that pretty quickly.

I'd love to see some shots of the elbow. I'm having a hard time visualizing that at the moment.

And I assume that by "singularities" you mean high-valence points or "stars" where 5 edges intersect?

vargatom

01-05-2011, 03:35 AM

A couple more notes...

- Always, always concept sculpt first! Even if you want to re-use base meshes, don't try to just model it without adding more subdivisions, hoping to get the right results just by pushing the existing few points... I've done it like this a few times: take the existing base mesh into zbrush, subdivide twice, rework it, then use retopo tools to fit the base mesh on top of the concept and add or rework edges where necessary.
This approach lets you make artistic and technical decisions independently, making for a better looking result and altogether a more efficient way to get there. It also allows you to completely rework the topology without loosing the sculpted forms; combined with the Transfer Attributes tool in Maya to preserve UVs and skinning info based on UVs, you can build a nice nonlinear workflow if the production schedules allow for that.

- Don't be afraid of denser base meshes. Looking at your model, I'd use at least 50 to 100% more edges because you can define more detail on the initial model, and you will find you need more edges to define detail that only appears in various poses.
There are many tools now to deal with dense models and the computers are fast enough, too. Most retopo tools have some sort of relax option so you only need to place vertices that should sit on sharp detail lines, like the concave borders between larger muscle groups, and you can then simply relax the rest of them.
I also use Daniel Pook-Kolb's PaintDeform tool for blendshape related stuff, it's the best thing ever:
http://dpk.stargrav.com/pafiledb/pafiledb.php?action=file&id=32
Or when you need to refit an existing and dense basemesh on top of a new sculpt, you can use a Wrap deformer; build a low-res version of the mesh and do the refitting with fewer points (much quicker) and wrap the final, highres mesh to the lowres one.
This helps you maintain the even spacing of the topology, and it's also good for heads - if you want to use the same topology, which I don't like, but might be necessary to keep bone weighting or re-use existing blendshapes... Still, as far as I can tell Blur's been doing this for a while, probably because of FaceRobot.

- Coincidentally, don't expect simple smooth skinning to give you proper results either. We almost always do at least a simple blendshape for the elbows because it's impossible to get proper results no matter how many influence objects you use. You need to shift the masses of the biceps and the triceps, add some skin sliding, compress the lower arm muscles etc.
If you're using blendshapes, you may need a special script though but I think either Jeff Unay's or Eric Miller's "Hyper real rigging" has one.
But the best approach is pose space deformation, especially with mocap because it tends to mess up joint rotations and you can't use them to drive the shapes. Michael Comet's poseDeformer is for free and some good souls always recompile it for newer Maya versions, for example. I expect I'll have to dig deep into it on our new project, so I might get back here with the experiences.

- Regarding the topology itself, Kel Solaar has already given some good advice like low number of poles/singularities/n-sided vertices, and nice example images. For a more realistic character I'd still fit some of the edgeflow closer to the general larger muscle shapes and deformation directions, particularly in the chest-armpit-back area. I would also increase the detail on the outer side of the elbows and knees - once you reach the maximum rotations on these joints, you'd find the existing detail insufficient to build up the bony shapes there.
I'll try to get some wire images myself, too, but right now everything is confidential because nowadays I only build stuff at work and that belongs to the clients...

BrettSinclair

01-05-2011, 04:48 AM

Thanks for the great info Tamas.

Kel Solaar

01-05-2011, 06:21 AM

Tamas: Brilliant post :) This Daniel tool looks awesome sauce, I saw it in the past but totally forgot about it, be sure I'll use it on my next shapes duty!

Tim: Yes, the singularities are >4 shared edges vertices (I call them Star Vertex).

Here is a small snippet to grab them (Singularities / Star Vertices) on the current selected objects:

import maya.cmds as cmds
cmds.polySelectConstraint(m=3, t=1, order=True, orb=(5, 65535))
cmds.polySelectConstraint(dis=True)

Another for NSides Faces if you are bothered to launch the Selection Constraint tool:

import maya.cmds as cmds
cmds.polySelectConstraint(m=3, t=8, sz=3)
cmds.polySelectConstraint(dis=True)

And finally Triangles Faces while we are at it :):

import maya.cmds as cmds
cmds.polySelectConstraint(m=3, t=8, sz=1)
cmds.polySelectConstraint(dis=True)

Just added them to my Snippets code repository on Github: https://github.com/KelSolaar/Snippets/blob/master/src/Maya/selectionConstraints.py

KS

El Burritoh

01-05-2011, 08:22 AM

Tamas, thanks for the insight! :notWorthy:

Kel, thanks for the snippets! :thumbsup:

I'm in the process of completely re-doing the pelvis area. I can't see how the one I posted here would allow proper deformation. Yuck!

I'm still a little stumped by how to approach the elbows and knees. If anyone has screenshots of the reverse of their characters, I'd love to see them! I guess it depends on the level of base detail...

The thing about "stars" is that I always see them, even on very high-quality models. Usually, the density is such that those stars don't make a real difference. But in the face, for example, I have never seen any example of "good" topology that didn't have stars somewhere on the mesh, in an area that isn't deformed too much. So even though they're carefully placed and avoided where possible, there's always a few. No?

Tamas makes a great point about base mesh density and modern hardware. My mesh is a little light primarily because once I add more geometry, that starts forcing me to define him too much. I specifically want to keep this model generic side. On the other hand....

...the other thing I've noticed is that if I start to define specific details and landmarks, it creates a mesh that looks good in that pose or in that level of "muscular tension." As long as those muscles move in only one direction it's okay. But as soon as adjacent muscle groups kick in to create an opposite motion, the underlying muscular detail has completely changed, and the "proper" topology for that is different from the one I had before. Which makes me think that indeed, high mesh density is the only way to allow arbitrary deformation without breaking. If that's the case, I want to avoid defining too many muscles, since that will need to happen with the rig or shape system.

I guess that's getting away from the generic and into the specific needs of a production then...

Kel Solaar

01-05-2011, 09:09 AM

The star vertices are not that much of an issue provided you try to limit them in areas with low deformation or surround them with enough faces to attenuate their impact. It's impossible to avoid them since they are the basis of edge flow direction change.

KS

El Burritoh

01-05-2011, 09:31 AM

That was my understanding as well.

I've been using Topogun 2 lately and it's really fun for exploring polyflow. At least for me it is... :D

Here's an update to the pelvis area. Just exploring options for allowing the greatest range of hip motion. I honestly don't like the some of the things I have here. I think this would allow some good range, but the flow is just awkward.

Does anyone model glute topology to follow the actual muscles as they wrap around from the coccyx and iliac crest to the front?

http://timcrowson.smugmug.com/photos/i-Pqk8jFM/0/L/i-Pqk8jFM-L.jpg

vargatom

01-05-2011, 10:03 AM

Yeah, stars are OK if you deal with them carefully; you're right that dense meshes make them stand out less, although I still have a few patterns I regularly use when I need to add or terminate detail, it usually results in a triangular patch of sorts, as I prefer to have as few of them as possible.
Oh and they're also disturbing because the Zbrush smooth tool cannot handle them; you can use Flatten instead but it's not the same.

Now this mesh is mostly good for sculpting, what you might want to do is simplify the back, the hands/feet, and the face/ear to some levels.
Generally the more you have detailed something in the base mesh, the more similar it will look in the end because you'll tend to skip resculpting the primary and secondary forms.
Faces are trickier though. In my experience sculptors do prefer to have the freedom to place the mouth, eyes and such wherever they want, but this usually also results in them skipping some work and cheating away the corners of the mouth and the eyelids, both very very important areas for building a good, animation-ready mesh. So it's up to everyone's preferences if they want to place rudimentary facial features on the base mesh or just keep it blank.

Also, once you start to sculpt characters of various body types, you'll either get faceted chests and arms, or overly dense areas on a thin limb which are hard to deal with. Not much of a problem for a concept sculpt, but it can be troublesome if you can't divide once again for a final detailed sculpt and displacement/normal maps become pixelated...

For deformations you'll always need to get a bit more specific to the character you want to create, IMHO. It also depends on how you actually plan to deform the character, as there are many possible approaches. I an't say much more now without getting lost in lengthy sentences, with english not being my first language... ;)

So in the end I personally don't really think that one basemesh is enough for every case. Building one for sculpting, retopo for the rest and trying to reuse as much as one can get away with is what I think to be a reasonable approach.

vargatom

01-05-2011, 10:26 AM

Does anyone model glute topology to follow the actual muscles as they wrap around from the coccyx and iliac crest to the front?

For the front I'd follow the line of the pelvic bone's top instead, both in the back and in the front; even on relatively flat people it tends to be there as a sharp line. Right now your edges are still running diagonally across this shape in the front.
Also make sure to add at least one ring of quad polys on both sides of this loop as well so that you can slide the skin above it when necessary.

For the back it's mostly okay that way IMHO, maybe redirect edges to follow the muscle fiber direction and/or fat deposits at the bottom? It's not necessary though, you usually don't get to see that part of a character, and the same goes for the thighs as well. When you do need to show definition you do have to trace the main separation lines or crevices between large muscle groups though.

6402

El Burritoh

01-05-2011, 10:43 AM

Thank you, Tamas! This is very helpful!

ulf

01-05-2011, 12:55 PM

@tamas
in z4 when you hold shift to smoooth, then realese shit while smoothing and smooth on, it changes to a different soooth algorithm that can deal with those too.

vargatom

05-05-2011, 10:00 AM

Infinite

05-05-2011, 10:39 AM

Ulf: I know that, but due to some bugs we're still using 3.5

By the way, I've just attended Ben Lambert's talk on creating Dobby for Potter 7 at FMX today, there's been some interesting stuff about body rigging and topology. I'll post a summary tomorrow (guess you guys are interested :) but I'm too tired to sum it up now, 2 flights within 24 hours and lack of sleep...

Lol :rofl: Glad to know I am not the only one who thinks that about ZB4. Most artists (and Pixologic, who don' listen) think I am nuts to still be using 3.5 but 4 has some nasty things going on.

Would be very interested to hear anything about stuff at FMX.

El Burritoh: I Wish I could help but topo is my weakest point. Tamas and Kel know their noodles! some great tips.

El Burritoh

05-05-2011, 10:40 AM

Don't get me started on ZB4. Such a bittersweet thing...

Tamas, I would love to know more about this Dobby character. Can't wait!

Infinite

05-05-2011, 10:52 AM

Don't get me started on ZB4. Such a bittersweet thing...

Tamas, I would love to know more about this Dobby character. Can't wait!

I agree with you, I had to open it the other day (reluctantly) and it felt sweet moving around, each update always feels different in movement and sculpting but I just can't use it here in my pipeline. Too many quirks.

Sorry to hijack, but just quickly. I'm interested to know others thoughts on Pixologic incorporating a UV viewer/manipulator etc and how others deal without the support for proper built in Multi UV map support ...and I don't mean multi UV islands.

Looking forward to seeing how your topo progresses. IMHO one of the hardest things to do. Especially full body.

tonytrout

05-05-2011, 07:44 PM

Informative thread especially regarding hips :thumbsup: I have a lot of questions but one I will ask :) relates to sculpting and retopologising - sculpting with mudbox on a base mesh I follow the advice of developers and use same sized polys so it subdivides evenly and you can add in detail where you want, but I get to a point where it becomes better to define some of the muscle mass demarcations with edgeloops otherwise my geometry becomes very heavy in areas I dont need a lot of sculpted detail in and if the edgeflow doesnt follow the structure like around an eye it becomes hard to form smooth shapes.

The question is that I rarely move anything to animation so my fingers are usually just even spaced sausages with maybe an extra loop at the joints, same with knees. I see quite a few meshes with extra squares on the outer of the knuckle joint or on the knee or elbow outer direction , is that advised for animation?

Cory

05-05-2011, 08:32 PM

For maintaining equal quads over the body, face, or any parts for that matter, I find this script to be quite helpful, especially with areas that get dense.

http://www.creativecrash.com/maya/marketplace/scripts-plugins/polygon/c/oarelaxverts

The script is for maya

BrettSinclair

05-05-2011, 09:35 PM

Interesting stuff here. Is this open for all to post edgeflow? I wouldnt mind getting mine ripped too pieces..
Vargatom: Very keen to hear bout that dobby presentation. Hopefully cgchannel post it up, or someone.

Infinite: Lol I know what you mean. I'm also on 3.5 r3. I've made a move to mudbox though for most of my sculpting needs. Just wish the mouse mode worked:(.

tonytrout

05-05-2011, 10:25 PM

Thats a pretty cool script Cory, I tried it out on one of my posed meshes and it ironed out some of the worst stretched areas. I use the relax brush in hexagon but it tends to collapse volume so this script is great in comparison. I see its recommended for blendshapes and I can see that would be useful. Thanks :thumbsup:

vargatom

05-05-2011, 10:56 PM

So, here are my notes on the body rigging of Dobby, then the rest of the talk.
Ben Lambert is a cool guy by the way, he's been at Framestore for a long time and made a very professional presentation that he said was his first. Lots of images, videos, notes, whatever, plus the final Dobby sequence from the movie which was very effective (no spoilers here - he apologized at the talk though ;)
The team was about 60 people and they've worked on both Dobby and Kreacher from september to may; Framestore got the job based on their previous work on nearly every Potter movie since part 3 and their Hyppogriff, and also after making a quick test in just one week.

They started with ILM's old maquette, which still exists; but there was also a prop for the actors to hold. It was quite heavy, and also one of the best paint jobs he's seen, so they've used it for SSS shader reference as well.
In general the design direction was about making Dobby less emaciated, more human looking, as he's now a free elf who doesn't beat himself up, and maybe he was also in a house elf spa since movie 2. They've smoothed out some of the wrinkles from the model they've rebuilt on top of the various scans (using Topogun).

In fact there were multiple models for the body.
They've made a medium res animation mesh that had anatomy (mostly the muscles) based edge loops to make the weighting easier, this was in order to better see how the body changes during animation. As there was no mocap, they wanted to keep the animators from breaking the model by showing them the masses of the shoulders, belly etc. They also had a skull in there, looked freaky with the red muscles and the eyeballs, like some zombie-Dobby.
Unfortunately no info on how exactly they did the weighting, I'd say simple Maya smooth bind is not enough for the results we've seen.

The final highres mesh topology was far more evenly laid out and had less anatomy features in the edge loops. But they had three separate versions of it, using the same topology and UVs: a bone mesh, a muscle mesh and a fat mesh. They've basically sculpted the same mesh to resemble a skeleton (obviously not segmented, but thin at the limbs, with rib cage and hip having more mass), then the musculature on top of it, and then the final skin with a layer of fat where appropriate.

The following part I wasn't able to take proper notes of but it goes something like this. They generate internal faces between the various layers to create volumes for the muscles and for the fat. They use ncloth in Maya to run simulations and they also do the jiggle deformations with these meshes.
The final skinning is however independent of this and done in a separate pass. They have their own solver to do the skin sliding and use it on top of the jiggle, Ben said it's far better and more efficient this way.

They also make heavy use of UVs and Maya's transfer mesh attributes tool to copy UVs between various versions of the same character model.
You can use UV matching to copy skin weighting and they have their own Wrap deformer which is based on UVs as well, so it's an integral part of their toolset.

As for the heads and faces, they've did some small changes here as well, smoothing out some larger folds around the mouth and smaller wrinkles on the eyelids as well. They also increased the separation between the brow area and the upper eyelid to make it read better - the maquette didn't have the deep fold of the skin that's usually there on most people. They were also worried about how the wrinkles would behave with eyelid motion and about possible intersecting geometry.
Ben said eyes don't just rotate, they also translate in the sockets when you look to the sides, it was important because Dobby's are very large and prominent so they couldn't skip it on him. They also used ncloth sims for the piece of tissue near the tear ducts, and some deformers to account for the cornea pushing the eyelids out as eyes aren't perfectly spherical either.

To make sure they get the eyes right they did an interesting test where they tracked the eyes of the voice actor on top of the first rendered pass of the CG character. Sounds creepy :) but worked surprisingly well, and gave them a very good example of what to do with the CG versions.

Reference was very important for everything else as well. They've even had a few expressions scanned from the voice actor, and they filmed a lot of themselves making expressions, even marking main wrinkles with black pen to see where the skin moves when relaxing their faces or using opposing expressions. They also filmed someone with UV makeup which provided a pattern that was easy to track visually to see the deformations. Ben said he thinks the proper eyebrow expressions are very important, and there's more than just the standard furrowing and raising (for example furrow and raise just the inner brow at the same time, it's a very different expression).

The face model was based on scans of the maquette, created in Mudbox, then modeled in Topogun. Overall it was a little less dense compared to the Avatar head models, but still quite heavy. Ben said it was better to have even topology for later sculpting work and blendshape modeling as the face had to accommodate various forms. Loops followed folds/wrinkles/deformation directions.

The rig was a combination of blend shapes and joints for the jaw and eyelids. Both joint systems used blendshapes on top for corrections, but there were no secondary bone/cluster based controls at all, only blendshapes for everything.
Shapes were based on FACS but in total there was about 150 of them, based on animator requests and probably including some of the automatic corrections. They also used many inbetweens, for about every 25% of the motion. They spent something like 4 months on building the shapes. Ben also mentioned that shapes for compressing and tensing the lips were important for stuff like making B/M and V phonemes.
They used the UV based wrap deformer to speed up the shape modeling, built a lower res mesh, wrapped the highres on top and finished in Mudbox. They also used animated displacements to increase the sense of tension, especially when lighting was flat for a start and SSS further removed a lot of the shading detail. They also had a special smooth deformer to maintain volume and flatten out wrinkles on the face selectively while maintaining volumes.

He also gave a little shot breakdown of the last sequence. They used a midget actor as a stand in for Dobby and also the puppet; they also made very detailed tracking models for the actor playing Harry and the others.
As there's significant body contact, they've used this tracking model and ncloth to deform Dobby's skin where hands touch him.
As much as I don't like Yates' work and directing style in the last 3 movies, I have to admit that this particular scene was still very effective. The VFX work was top notch as well, animation, shading, compositing, and the talk gave a good insight into the production process.

BrettSinclair

05-05-2011, 11:21 PM

That's not a job for one man >_<. Thanks for the info. Did they mention if they modeled tear ducts separate, as they applied an ncloth to them. Also interesting about topogun. I thought cyslice was more common for the retopo job. Thanks again :)

vargatom

05-05-2011, 11:28 PM

Tearducts were a separate piece, yes, and quite large to cover the entire range of the eyeball rotations.

I'm not that surprised about Topogun; cyslice is an ancient piece of software by now, lacking in modern features and has a terrible GUI. It was originally developed to resurface digitized maquettes with NURBS patches, back in the late '90s.

Ben Lambert was head of modeling, they also had another modeler for Kreacher, a texture artist, and four riggers; I think the rest were animators, compositors and tracing/roto artists.

If anyone's still in Stuttgart or can get there, they'll have a roundtable discussion on virtual humans later today, with guys from Pixar, DD and maybe ILM I guess... I had to get back to work though.

Oh and there was also a talk on the original Tron by one of the animators. He even had the graph papers on which they've calculated coordinates for camera and lightcycle movements, as well as all the storyboards he made for the game sequence. I think it had the most attendees and the biggest applause on the entire conference, although I haven't had the chance to see stuff like ILM's Rango talk. FMX is really like a somewhat smaller Siggraph for Europe I guess, I wonder why I didn't hear about it before...

BrettSinclair

05-05-2011, 11:45 PM

Interesting about the tearduct. Might be over kill though. Surely a blendshape or something would do the job. Looking at neytiris face wires it seems they built those right in. Easier to keep track of them that way. I suppose every house has its own rules..

vargatom

05-05-2011, 11:49 PM

Dobby's eyeballs are so large both absolutely and in relation to the face that blendshapes don't work as well as on human faces. Even the area of the iris is far larger compared to the entire eyeball than what you have with us humans.
That's probably why they went for joints for the eyelids, to keep the vertices on top of the eyeballs; and for the jaw as well, the entire head is so pointy and elongated that the linear transitions of blendshapes are just not enough.

I personally get good results with shapes on human characters, too, but I can see how it'd break down on something like Dobby, even with inbetween shapes.

BrettSinclair

05-05-2011, 11:51 PM

Ye that's true. Good point..

rasmusW

06-05-2011, 07:01 AM

this thread is such a good example of why i like this forum so much..
this is a fantastic read and thanks tamas for your notes on the dobby talk.

-r

blambert

07-05-2011, 01:13 PM

Hi

Many thanks for your interest in my talk, I saw a link to this post. Its encourgaing that you found my FMX talk useful.

FYI, the tear ducts (caruncaulars) were not done using ncloth. They were modelled in a neutral pose slightly folded up, then, with some careful rigging they would fully 'unfold' flat as the eye rotated or moved around. All that was done using weighting by rigging. Because the eyes were so big, we had to go this far, the tear ducts really do move a lot when the eyes move around.

Ben Lambert

BrettSinclair

07-05-2011, 09:34 PM

Hey Ben thanks for clearing that up. Glad to hear it was rigged. Is there any chance we can see the wireframes for the different versions of Dobby? Btw I think Dobby and Kreacher looked the best in the latest movie. Eye popping performance from you guys. Congrats.

PascalR

07-05-2011, 11:21 PM

Great thread guys, I hope Ben can share a few Dobby screenshots, I'd understand if that's not possible though, can't say much about our own pipeline:o
Tamas, great posts !!
:sw:plugged:sw:

kenshiro

07-05-2011, 11:50 PM

I was at the FMX this week , and I saw the Harry potter Tribute, it was really intersting, awesome work on Dobby and Kreacher from ILM and Framestore, I really enjoyed this conferences with all studios, an awesome work!

Infinite

07-05-2011, 11:50 PM

Really cool thread, thanks for the extra info Tamas and blambert. Really informative.

vargatom

08-05-2011, 12:22 AM

Heh, thanks Pascal but the credit should go to Ben and the Framestore team, I've only reported about it :)

Ben, thanks again for the talk, it was very informative indeed. It confirmed to me that we're going in the right direction with our own stuff, and I'm still getting new ideas on what do do next. Hope it's okay with you that I've posted my notes to share it with others here!

Tim, I've just found this video online from a past interview, it includes a few frames of the collision mesh we used for the cloth sims, maybe it can help a bit:
http://www.syflex.biz/clnt/a2_ezio.mov

LarssonGu

08-05-2011, 04:58 AM

Awesome thread, all last year students from my school were at FMX. Me and some guys begged our cordinator to let us go with them, but we're working on a large project so we weren't allowed D: Seems like everyone had a great experience there! Gonna hear about it tonight!

Major applauses to Framestore for the great work. Was stunning and insanely encouraging.

cheers for sharing the great notes!

3mm

10-05-2011, 12:04 AM

great thread,thanks so much Ben And Tamas for recording great notesclap

El Burritoh

11-05-2011, 03:54 AM

Tamas, thanks a million for sharing your notes on Ben's talk. It's great to have you here, Ben! I'll echo Pascal's hopes for some examples.

Brett, by all means! Please post your topology! Having a variety to examine can only help.

I've been swamped lately but I'll be re-examining the pelvis in the example I posted previously.

Ben, it's great to know you guys used Topogun so much. I've been having a blast with v2.0 . To me, it seems like an ideal way to explore different topologies quickly.

Mikademius

11-05-2011, 04:14 PM

Hey El B.
Great work so far. I have a couple of notes. I don't know if it's gonna give you any problems really, but I thought I should point it out. You currently do not have any direct connections between the eyes and the mouth. If you look yourself in the mirror when you sneer, you can see how the muscles pull the corner of your mouth UP.
Also, I'm wondering if that tri at the thumb will give you any issues? There is a lot of deformation right there. I hope this tread will be active because it's a very important part of our day :)

Take a look here for some facial topo I find very good:
http://www.sintel.org/gallery/

El Burritoh

11-05-2011, 04:28 PM

Thanks for the input, Martin! That's an excellent observation about the eyes to mouth area! I'll be sure to add that to my list of changesq and get it cleaned up for the next posting. Likewise, the triangle in the thumb is sloppy too. Could be tricky to fix but I'll check it out.

vargatom

11-05-2011, 10:31 PM

Facial topology is a completely different issue IMHO. The first example I'd study is this thread from Leslie, even though there are a few very minor issues about the model (mainly the eyebrow area right above the bridge of the nose), it still works wonderfully.
http://www.cgfeedback.com/cgfeedback/showthread.php?p=15500#post15500

Or there's Neytiri:
http://img62.imageshack.us/img62/3059/neytiriwiresmall.jpg

Of course you also have to keep in mind that these are young female characters with soft facial features, the more angular or older the face gets, the more different the topology has to be so that it can accommodate both the static forms and all the variously distorted versions for the blendshapes used in facial animation.

BrettSinclair

11-05-2011, 10:52 PM

Seen these neytiri wires before. After seeing it I cant understand why people freak out over stars.

vargatom

12-05-2011, 03:54 AM

People who are worried about those are usually building head models of 1000-2500 polygons, whereas Neytiri is about 20-25k. There's a big difference in surface continuity between the two; although you also have to keep in mind that skin details usually make the surface very bumpy so it's not such a big problem in the end.

BrettSinclair

12-05-2011, 04:11 AM

But would a topology like that be useful in other facilities? I would assume people would run for the hills. I'm all for modeling stuff in, I think its got more control. But where is the cut off. haha.

vargatom

12-05-2011, 04:29 AM

Well, we haven't seen any major increases in production times, I guess it's because you tend to worry less about terminating edges, you just cut through the entire model instead. We already have tools like soft selection, wrap deformers and Daniel's paintDeform brush tool to manage the more detailed geometry, I guess most studios have been on this path for a while now, we just didn't know about it.

Most of our characters are 100-250k polygons nowadays and of that, 8-10k is the head. The only thing you might worry about is file sizes and interactive speed of the deformation rigs, for example while painting weights... but I haven't heard any complaints and the lead rigger sits right in front of me :)

Kel Solaar

12-05-2011, 04:48 AM

BrettSinclair

12-05-2011, 06:05 AM

Well I thought I'd put my work on the chopping block. I've been told this is to dense but from what you guys are saying its low poly:) I'd like to add in more stuff for crows feet and brow wrinkles. As well as more chin detail..To name a few, I also tried a different approach for the eyelids instead of the normal looping.
6553.

Jake_H

12-05-2011, 11:05 AM

Good info regarding resolutions on heads for production -
I always assumed people kept stuff relatively low for skining and rigging purposes and then drove the surface details it with shaders - perhaps thats a too games centric approach.

PS when I say low I mean as "low" as Bretts post above.

Infinite

12-05-2011, 01:34 PM

vargatom

12-05-2011, 02:29 PM

El Burritoh

12-05-2011, 05:07 PM

Looking at all these recent posts makes me want to rebuild my guy from scratch using a higher density mesh, just to make sure I get the main features in place with greater accuracy, and to get more detail in the face. My mesh is definitely "low-poly" by these standards. But then again, I'm really just wanting something I can pose in a 3D app and then sculpt in Zbrush.

It's kinda comforting to know that modern productions are all using higher density meshes. It means we don't have to rely on displacement quite as much.

BrettSinclair

12-05-2011, 07:23 PM

tonytrout

12-05-2011, 08:03 PM

Wouldnt you have a number of different resolution meshes for different shots, with different texture and displacement resolutions and number of maps. I did read somewhere that there were a ridiculous number of textures on the trex from king kong so it must be difficult to pull these in and out for rendering

On the neytiri topology I notice the outer eye corner is on a crossover bisect rather than a loop, something I have been trying lately, Also I like the way the mesh resolution has been reduced as the eye socket heads towards the scalp

oglu

12-05-2011, 08:29 PM

my last retopo work...
there are some areas on the forehaed i like to change...
what do you think..?

6555

PascalR

12-05-2011, 09:40 PM

It would be really nice if some one from weta could chime in here and clear this debate over the subdivs up,It has been plastered all over the show so I dont see why not. Infinite's paint over would suggest a smooth subdiv is used, Or perhaps smoothed and deleted? Then again looking at the eyes area who knows.
Can't say anything :D Lots of polygons were sculpted on Neytiri's face by Master Syomka, he would be the best person to come to talk about this here and lose his job right away. :rofl:

BrettSinclair

12-05-2011, 09:42 PM

Can't say anything :D Lots of polygons were sculpted on Neytiri's face by Master Syomka, he would be the best person to come to talk about this here and lose his job right away. :rofl:

Hahaha. Understandable. So can I get an invite to come view it then? :sw::D

Kel Solaar

12-05-2011, 09:57 PM

Actually what Mister X and Mister Y did when we worked on Project Z was to do the blendshapes on the head at default level so that moving masses was easier, then they smoothed them and proceeded to add more details and subtlety. I then added animated displacement over that.

They however never started the shapes at smoothed level. Mister X was coming back from Weta (Avatar Preproduction) at that time.

If you know them, please don't write their names here, thanks!

KS

Infinite

12-05-2011, 09:57 PM

That's probably how the model was made, but there's been info on another forum that this was indeed the 25k polygon basemesh, used to sculpt all the expressions so that it will work without any animated displacement maps. The big Making of Avatar book has some images of Jake's avatar, and one is closeup with some blendshapes dialed on, and there's also the behind the scenes movie on Yahoo from which I've grabbed these images, also showing Jake's blendshapes (although without wire overlay).

I agree but I think it can be misleading to some people when they see shots like that as they might presume you need to build at that resolution or near to it which is plain crazy.

On that paint over it could even be possible to go down another notch.

I think the build of the model is low but really well structured, then as you say perhaps they use the highest subdiv level for rigging and deformation, and a mult-res model for sculpting, which would make more sense.

Shame we will never know (well maybe in 5-10 years time) dash those secretive masters at Weta!

Tamas: Just saw Digic's Mass Effect 3 cinematic again the other day. That main guys face is incredible. It has that Tron type quality to it, high end but up another notch with no creep factor at all.

Brett and Oglu cool meshes!

Kel Solaar

12-05-2011, 10:05 PM

Shame we will never know (well maybe in 5-10 years time) dash those secretive masters at Weta!

Actually Lee we cross posted :) I would suspect it was done the way Mister X and Y did it on Project Z. (Refer to my post before yours) :)

Infinite

12-05-2011, 10:11 PM

Actually Lee we cross posted :) I would suspect it was done the way Mister X and Y did it on Project Z. (Refer to my post before yours) :)

Sorry that is way to cryptic for me to understand :rofl: like Algebra. Does it mean the answer = D?

Storm

12-05-2011, 10:17 PM

This might seem like a really stupid observation but is it not possible those wire shots of Neytiri aren't showing the lowest subdivision level?

Excuse the naff PO:

http://www.ir-ltd.net/uploads/weta-eg-02.jpg

I'd say your paint over looks like the actual level, it works with the ear as well. Couldn't have been any lower though, then the ear topo wouldn't work.
6556

Btw awesome thread guys, so much great information!

tonytrout

12-05-2011, 10:22 PM

I then added animated displacement over that.
KS

Kel could you describe that a little more? Is it animating the displacements in sync with the blendshapes?

LarssonGu

12-05-2011, 10:32 PM

Exiting, I always thought it was modeled that high rez, never thougth of doing a paintover. But ye, watching some of the Weta people model stuff like this would certainly be a real treat!

vargatom

12-05-2011, 11:49 PM

I agree but I think it can be misleading to some people when they see shots like that as they might presume you need to build at that resolution or near to it which is plain crazy.

Noone has to do anything :) It is just one possible approach to a complex problem, there are other options as well.
The Tron guys used a lot of animated displacements for nearly all the wrinkles. Apparently they also had a lower resolution version of the same head, if you watch any of their behind the scenes stuff you can see that one with the wireframe overlay shots.
The Framestore guys working on Dobby built a low res mesh on top of the final one to rough out the blend shapes and wrapped the highres mesh to it. (I've been doing something like this for years, too)

I think the build of the model is low but really well structured, then as you say perhaps they use the highest subdiv level for rigging and deformation, and a mult-res model for sculpting, which would make more sense.

It's still a mesh with about 6-7 thousand polygons before the last subdivision, so it still takes some modeling effort to place those edges nicely and line them up with the facial features. I imagine there's some concept sculpt involved in the process at some point as well, but maybe they don't do it with a retopo approach...

Tamas: Just saw Digic's Mass Effect 3 cinematic again the other day. That main guys face is incredible. It has that Tron type quality to it, high end but up another notch with no creep factor at all.

Hey, thanks a lot, although I'm a bit unsatisfied with the results, there wasn't enough time to finetune it the way I would've liked to. Most of the credit goes to the shading and texture artists, and the lighting/comp; the facial rig could've been better, more polygons to provide better deformations and wrinkles... Maybe next time, we finally have a schedule with some more room for this stuff ;)

I'll check the mesh screenshots later today, they look good at first glance though.

BrettSinclair

12-05-2011, 11:55 PM

Hey Tamas, Any chance of seeing some topo from you guys and what you expect from a model? I've watched that Mass Effect 3 teaser a couple times as well. Its really sexy.

El Burritoh

13-05-2011, 01:32 AM

my last retopo work...
there are some areas on the forehaed i like to change...
what do you think..?

6555

Nice work!

The only thing I wonder about here is what Martin mentioned about mine: does not having a clear loop connected the lower eyelid to the top lip mess anything up?

Do you have the rest of the body too?

vargatom

13-05-2011, 01:39 AM

Well, it's unofficial so I can link to it, it's been photographed here when a local news portal did an interview with us on some related topic.

http://galeria.index.hu/kult/2011/01/21/digic/1902041_703eb933540a07a9b2944eaf4860ca35_xl.jpg

We're doing things differently now - eyebrow and neck have different topology, and the rest is a little more dense, particularly where the crow's feet wrinkles go and on the sides of the nose.
This is about 6000 faces, we're above 9000 with current human characters (we have completed assets for three more movies since ME3).

oglu

13-05-2011, 02:14 AM

Do you have the rest of the body too?

not yet... but id like to do the upper body similar to the mari blacksmith...
specially the hands are really good there...

Infinite

13-05-2011, 06:44 AM

Kel Solaar

13-05-2011, 07:22 AM

LarssonGu

13-05-2011, 07:54 AM

ye, it looks awesome! Love it when people are able to share this stuff. This forum becomes better and better every day.

BrettSinclair

13-05-2011, 08:14 AM

Thanks for sharing this Tamas. Are you guys creating new heads from scratch every time or just improving on the current ones? Sorry for all the questions :o

ulf

13-05-2011, 09:40 AM

really great thread! thanks for sharing to all!
@tamas - your cinematics are just amazing!
@brett, cool mesh and character.
maybe only the mesh flow around his neck could go around straighter?
you plan on or already do blendshapes for him?

vargatom

13-05-2011, 10:10 PM

Hey guys, you're welcome :) I actually have a few other images still waiting for client approval but now that I'm out of the E3 crunch I'll get them out as soon as I can.

Brett, we're usually starting from scratch - even though there are similar topology patterns on the heads here and there, it's ultimately still a unique model for each character. It works like this for us, but again, others might find different approaches better.
For example, I think Blur uses the same mesh, probably because FaceRobot involves a lot of weight painting and they can probably reuse the results there to save time. Naughty Dog has the same head model for their characters as well, because their ingame face rigs are all joint based. I guess if you use something like a wrap deformer and a very lowres wrapper model, you can refit a head mesh to a new face; then clean up the details using any retopo tool, and get it all done within a few hours.

BrettSinclair

13-05-2011, 11:49 PM

Ulf: Ye I could make the loops straight on the neck but I wanted to implement the sternomastoid. I could do a better topology though:). Your model is looking very clean. Good job. I would perhaps just look at the ears and eyes. The eyes seem to have been pulled up like a face lift almost and the ear could use some more attention on the anti-helix fold. Though there are probably a lot of ears like that so its minor.

Tamas: Thanks for the reply. So do you guys do a base concept sculpt or scan then retopo? I'd assume this is the best approach. Also are you guys using a hybrid face rig? Do you create a new rig for every head? Sounds hectic but I suppose then at least you have specific tools for the model. Looking forward too some new images. Perhaps some new renders of the soldier in the cine would be great too see:)..

vargatom

14-05-2011, 12:18 AM

Well, we do concept sculpts, or sometimes use the Zbrush models from the client when they have one for their ingame character.

We only use blendshapes in the rig, and have the control node network as a separate asset that takes a single click to add to the scene and another two to attach to the head mesh with the blendshape node. Only nested blendshapes need some extra work to set up but it's still quite streamlined nowadays.

vargatom

14-05-2011, 12:29 AM

Oh, yes, nested blendshapes:
basically you can chain blendshaped heads together in Maya. It's like Inception, a blendshape within a blendshape ;)

Say you have a target for opening the jaw that also opens the mouth.
Then you make another target that's an opened jaw but with the mouth kept closed. You add this to create a blendshaped version of the jawOpen target.

jawOpenMouthClosed -> jawOpen -> final head

This way you can add an extra control to change how much the mouth opens when you open the jaw. Usually when people start to talk the jaw starts the movement a few frames before the lips open and this helps you animate that. Or you can make chewing movements or anything like that.
I also use this to do sticky lips but it's a bit harder to explain... basically the low level blendshape has a set of inbetweens that pulls the lip vertices together starting from the outside. So this controls the width of the sticky effect, and then the main blendshape controls the amount that you have to set manually, based on how far apart the lips are.
It's not as accurate as the Hyperreal rigging DVD's version which uses clusters attached to the surface and driven keys; but it's very quick to rig and doesn't require anything else but blendshapes. We're looking into some custom deformers though, to see if those would work better.

So this nested blendshape stuff is a strange concept at first, I think I've gotten the ideas from some posts by Jason Osipa somewhere. And I'm sure I've only scratched the surface of what else can be done with it...

ulf

14-05-2011, 02:02 AM

hey cool, interesting that nested blendshapes and sticky lips setup, thanks for that tamas!

BrettSinclair

14-05-2011, 02:16 AM

Thanks for the explanation:). Sure wish I had some guy to rig my stuff so I can play :D

vargatom

14-05-2011, 03:10 AM

We use Judd Simantov's jsFacialWin.mel script to build the controls and assign them to the various shapes and attributes. It's quite easy to use IMHO; it only gets complicated when you want multiple controllers to drive the same shapes, or add joint rotation based controls to some of the shapes (like eyeballs driving eyelid deformations).

The way we make the resulting control node network reusable is to have a simple locator with attributes sit at the end, and connect this to the actual weights of the blendshape node. This way you can always replace either one of them; edit the rig itself in a separate file and reimport it, or rework the mesh and the shapes and still keep the rigging. A simple script can run through the attributes and build the connections, it's like ten lines long or so.

collings

14-05-2011, 06:06 AM

great thread :) thanks for all the interesting information.

For example, I think Blur uses the same mesh, probably because FaceRobot involves a lot of weight painting and they can probably reuse the results there to save time.

We don't actually, some years ago they use to recycle the same basemesh but nowdays each head is usually pretty much unique depending if it's an old guy the mesh will be much more heavier topology wise than for a young lady to support the wrinkles deformation .
Also we are using other technique sometimes than FaceRobot when it comes to animate realistic face. (MOVA capture, Blendshapes, etc... ) .

BrettSinclair

16-05-2011, 07:10 PM

Vargatom:Ye I used the Judd script before(Long ago) was super cool and easy. But the rest you mentioned I've never done. Been ages since I rigged up anything.

Collings: Hey Nic, nice to hear from blur here too:)

BrettSinclair

17-05-2011, 09:45 AM

Hey guys. I was wondering if anyone knows of a script too mirror blendshapes on an asymmetrical model? Thought I'd just ask in here instead of starting a thread:p

Thanks

Kel Solaar

17-05-2011, 10:06 AM

Brett: I don't have anything coming to my mind, and for what Software? :D
If the model is not too much asymetrical and you do not plan to do asymetrical blendshapes, maybe you can symetrize it, do your blendshapes in a symetric fashion, and add the asymetric head as a new target.

KS

BrettSinclair

17-05-2011, 10:09 AM

Hey Thomas thanks for the quick reply. Its for maya, I found one script called edgeFlowMirror but it has a dynamic loading error so I assume its 32 bit only(death to 32 bits,lol) http://www.creativecrash.com/maya/downloads/scripts-plugins/modeling/poly-tools/c/edgeflowmirror/description#tabs. It looks/sounds really handy. Couldnt get it working though.

ulf

17-05-2011, 10:30 AM

saw this and thought i post this in this thread.
wow
http://www.youtube.com/watch?v=pZTN-jRRM30&feature=player_embedded

vargatom

18-05-2011, 12:44 AM

I'd try wrap deformers for mirroring the shapes; you duplicate your original head, add the one-sided blendshape, then you apply a -1 transformation. Now the shape moves vertices on the other side and you can wrap the original head to this mirrored one, then combine it together with the one-sided shape (like, add another blendshape deformer and mix them together, or paint a mask). I do this in Max and it works reasonably well... if you have some serious asymmetry, you can try to move the mirrored head around a bit before the wrapping.

vargatom

18-05-2011, 12:46 AM

Ulf, there's some good stuff on that head, but quite a lot of not so good stuff too, IMHO... The fading in/out of the displaced brow wrinkles in particular is quite bad, and many facial deformations don't have a proper falloff IMHO. Also the lipsync at the end is very stiff.
But many times it does look like there's muscle and bone under the skin, so maybe it just needs some more tweaking. But I'd definitely lose the displaced wrinkles, especially as it seems that they have enough edges to sculpt them into the blendshapes...

BrettSinclair

18-05-2011, 12:57 AM

Thanks Tamas I'll try that out:)

oglu

18-05-2011, 06:25 AM

update on my male headmesh...

6657

vargatom

18-05-2011, 09:17 AM

Do you plan to animate it, using blendshapes or such? If that's the case I'd have a few suggestions but I don't have Photoshop at home to do an overpaint right now...

oglu

18-05-2011, 09:21 AM

would be great to get an overpaint... id like to do some blendshape tests with him...

you could use mypaint... :D
http://mypaint.intilinux.com/?page_id=9

vargatom

18-05-2011, 10:02 PM

Here are some comments, sorry I went to sleep yesterday and didn't see your reply ;)

http://imageshack.us/m/832/5100/overpaint.jpg

BrettSinclair

19-05-2011, 04:31 AM

RasmusW showed me this today and thought I would share it. I droool.. hehe
http://www.chadvernon.com/blog/resources/cvstickycurve/comment-page-1/#comment-780

oglu

19-05-2011, 08:24 AM

update... need to even out the spans a bit but the topo should work now...
what do you think..`?

6675

collings

20-05-2011, 07:01 AM

hey oglu, is it suppose to be your head basemesh for sculpt , or is it suppose to be a final head with good topo ready for animation ?

because i find it always way better to actually sculpt first your head and then depending of where and how you sculpt your wrinkles you lay down your final topology accordingly to it. That will give a much more unique look.

i agree with what tamas pointed out in the paintover. I see that you just changed the eyebrows and mouth area. Maybe you should also adapt the other area like the edge direction underneath the lower eyelid for proper wrinkle deformation.

In addition to Tamas feedback, you should put more edgeloops and forms in the upper eyelids. As it is now, you ll have trouble to make a correct blink shape.

zeoyn

20-05-2011, 07:08 AM

What a great thread!
I have been following this closely and learned a lot, so I have decided to contribute by sending the topology of a base head I have just created to be used as base for sculpting. I have tried that the edge flow follows the muscle groups as much as possible.
Tips for improvement are more than welcome!

6692

oglu

20-05-2011, 07:28 AM

hey oglu, is it suppose to be your head basemesh for sculpt , or is it suppose to be a final head with good topo ready for animation ?

Maybe you should also adapt the other area like the edge direction underneath the lower eyelid for proper wrinkle deformation.

for the moment it should be both... ive done a quik concept sculpt for this one...
but i drifted away... its my first head in this high polycount... so i like to finish this one with a second sculpting session...

ive changed the eye area a bit more but only moved the loops upwards... ive not cuted in a new direction...

if this is not enough could someone do an overpaint...?

oglu

20-05-2011, 08:23 AM

next one is a girl...
still working on the lips...

but i have here the same problem in the eye area...

6694

Kel Solaar

20-05-2011, 10:15 AM

Infinite

20-05-2011, 12:11 PM

Here is an old but interesting paper about Uncharted 2 Characters, posting here for reference:
http://cmpmedia.vo.llnwd.net/o1/vault/gdc10/slides/simantov_judd_uncharted2_character_pipeline.pdf

There are some good ideas, like building the characters with the mouth a bit opened so you can have clean corners out of the box (This is important when you are doing mouth blendshapes). I started to use that trick on my characters :)

KS

Smart! :thumbsup: Sounds ideal, never thought of trying that. They are the hardest thing to work with when sculpting I find. Give me an ear any day!

I can't offer any guidance on this thread, I am free riding here, please excuse me. These other guys really know their beans!! Hope to see some more production type screenies.

Grandizer

20-05-2011, 05:27 PM

Tamas - Thx for sharing the head topo - work you guys do is awesome
- nose to mouth crease/folds looks perfect ... if you ever get to post a close up of eye area would be sweet :D

Olle

22-05-2011, 10:31 PM

Here is some more information on Dobby and Kreacher:

http://www.3dworldmag.com/2011/05/13/framestore-reimagine-dobby-for-deathly-hallows/

Cory

22-07-2011, 11:43 AM

Probably ages old but my first time seeing this. lol

http://img829.imageshack.us/img829/2015/ferfp.png

El Burritoh

27-07-2011, 05:24 AM

Hippydrome posted this video on Vimeo a few days ago. This is a very instructive video on his thought behind facial topology:

Can't embed it here, but it's a must-see.

http://www.vimeo.com/25969681

Cory

28-07-2011, 03:48 PM

Thanks for the link tim, very informative video.

andreicirdu

01-09-2011, 02:28 AM

http://www.danplatt.com/ Dan Platt posted some of his head models from the TRON movie.
He posted Clu's and Quorra's wireframes http://www.danplatt.com/?p=194 http://www.danplatt.com/?p=151

oglu

01-09-2011, 02:58 AM

thanks Andrei..!
Dan is an great artist... i like his stuff..
the meshes are lower than expected... looks like a lot of displacement...

Infinite

01-09-2011, 04:55 AM

http://www.danplatt.com/ Dan Platt posted some of his head models from the TRON movie.
He posted Clu's and Quorra's wireframes http://www.danplatt.com/?p=194 http://www.danplatt.com/?p=151

Fantastic links! thanks for sharing.

3mm

01-09-2011, 08:44 AM

Thanks for sharing Dan's works,Awesome stuff:drool:

phrenzy84

01-09-2011, 09:53 AM

thanks Andrei..!
Dan is an great artist... i like his stuff..
the meshes are lower than expected... looks like a lot of displacement...

Yeah i imagine it would be hard to intuitively hand animate that stuff.

Amazing modelling skills.

Im working on a highly detailed retopologized mesh for ultimate control. Would love to know what you guys think of it since that kind of stuff hasnt been posted here yet. Im finishing up the head but already the body along weighs 15k polys (if i remember correctly). I guess it will require less displacement. :)

j3st3r

13-09-2011, 10:34 PM

Great topic guys, learned a lot. I'm going to start on a new character, so I can use the above mentioned. However, I'm quite interested in topologies at shoulder, and at joints, like elbow and knee, outer and inner areas. Many of us are single artist rig our own models, without any technical artist, avoiding the use of nested stuffs, blendshapes driving skinned models, etc. What could be the best for the deformation, and the shape?

Cheers

Jester

phrenzy84

17-09-2011, 07:47 AM

Ive been doing alot of research in this area and although its not complete and far from perfect (which i think isnt possible :)), thought i might share a piece i have been retopologizing.

Im obviously going for the high poly base, for alot of control. There are flows in the model that arent the norm simply because of the model but this is a relatively old chap. If he was younger then it would be much less in poly count and much 'smoother'.

https://lh6.googleusercontent.com/-OBAhR6u67Xw/TnN2qgN4IBI/AAAAAAAAAgs/tNxpVxT_Iy4/s800/frontBackTNs.jpg

https://lh6.googleusercontent.com/-ZhKR9Zcp9AM/TnN2f3XHxRI/AAAAAAAAAgc/XwVPkhJQw2w/headtopoTNb.jpg

Like i said its isnt finished so there are some poles and patches to clean up.

MDB101

23-09-2011, 08:33 AM

Hippydrome vids are gone D; but the website is still up ;D

http://www.hippydrome.com/

Rox

23-09-2011, 11:28 PM

Hi. Here's the body for a personal animation I'm working on. The body will be used as a cloth collider. Please tell me what you think about topology, because I'm not quite sure it's correct or not. Thanks

8847

Time2Time

04-10-2011, 12:42 AM

nfuchs

26-10-2011, 08:52 AM

I was just trying to build a simple nested blendshape rig as Vargatom explained (thank you for the info!), but in Max. It didn't work. Then I found that there is a 'Progressive Morph' area - on the Morpher modifier's panel - which can be used to create such things as sticky lips and so on. Basically you just add all your mouth targets (lips opened, lips sticking together) into the same channel and they will show up in the 'Progressive Morph' list and you can adjust how much influence the targets have individually. Just thought I share this in case you are working in Max to achieve sticky lips, etc...

phrenzy84

11-11-2011, 07:29 AM

https://lh6.googleusercontent.com/--KfXs1MQ63s/TrwYnc2al_I/AAAAAAAAAiw/fv4yTYaogOU/s800/morph_Test_TN.jpg

https://lh6.googleusercontent.com/--8ThN7kF2w0/TrwaP_dXQiI/AAAAAAAAAjE/jVLsDLCepeg/s800/morph_TestWireTN.jpg

Posted this in the Topology thread over @ cgtalk thought i should also post the update here.

It's just a quick pose test i did with my revised topology of the old man.
Please note these are just simple rotations so no volume has been preserved or removed from both the poses, also there is no corrected areas where there is pinching .

3mm

26-01-2012, 04:28 PM

Just find it and i think its the same presentation which Tamas talked about:
http://www.youtube.com/watch?v=hNKHa5hyHbU

rasmusW

27-01-2012, 01:13 AM

thanks mohammed!

this was fantastic.

-r

vargatom

08-05-2012, 02:23 AM

Hey guys, there'll be another session on virtual humans at FMX tomorrow, talks from Double Negative and Weta Digital. Also, more talks on John Carter and Avengers will follow on Tuesday and Thorsday.
I'll be there ;) so if anyone wants to meet up after hours for a beer, PM me or email or something.

rasmusW

08-05-2012, 02:53 AM

arhhh!!! sounds fantastic and very inspirational, tamas...
would love to experience fmx one day.

since i ain't attending myself, one can only hope for such fantastic notes as last time;)

-r

vargatom

08-05-2012, 03:16 AM

Well I packed my notebook ;) I think that taking pictures is not allowed though.

Rox

08-05-2012, 04:21 AM

Hey guys I have a question. When I have a hi-res character model, should I use the same meshes for the animation process(bind with joints etc)? Or create a simplified version for animation to speed up the workflow and then transfer the animation data from the low res to the hi res via deformers?

Also how do you deal with tricky body parts like armpits, shoulders and such? I mean, sometimes a character moves his arm very close to the body and it interpenetrates with it. It's okay with very simple character, but when you have simulated clothes and moving armor elements everything literally explodes. Do you somehow simulate collisions on the body or use muscle collisions there?

thanks

Adam Baroody

09-05-2012, 03:29 PM

I'm working on a spider-man model for fun, and I've actually run into an interesting problem with topology, and would be curious to hear some thoughts on it.

Basically, I'm making the spider-man body suit. It has seam lines and panels that flow along with the design of the costume. Now obviously the seams lines are not very high, and they could easily be handled by a displacement map, that is applied to a simple flowing base mesh.

However as per the discussion in this thread, I too subscribe to the idea of having a higher polygon basemesh to control the shape deformation. (wrinkles and things as the suit bends) However I'm not sure if i should model the topo to follow the suits seams or not. The base mesh will be high poly no matter what, so i can control wrinkles etc, but I'm curious to hear everyones thoughts on which method they would use, the pros and cons etc. Obviously they are both viable choices, but I seem to be leaning towards the topo following the suits seams and panels. Especially because it benefits areas like the gloves, especially the palm features which have their own fabric seems and details.

The only problem though is, when sculpting shape deformation wrinkles and things, it can become very easy to smooth out the seams, and essentially erase them. So it takes a lot of care and even re-sculpting to bring back those seams after sculpting the base geometry.

Below is an example of the way the different topo flows on the mesh. The final image is a test deformation of the arm, to see how the topo would deform with the edges flowing along the fabric seams.

Any thoughts would be appreciated.

http://adambaroody.com/dump/spidey_topo_problem.jpg

vargatom

11-05-2012, 10:57 AM

Adam, you should try Daniel Pook-Kolb's paintDeform script, it should help bringing back the modeled seams.

http://dpk.stargrav.com/bcs/docs/data/lv2-paint-overview.html#paint-overview

vargatom

11-05-2012, 11:03 AM

So I'm back from FMx, which was quite good this year as well even if I had a fever for the first 1.5 days. I've attended talks on CG characters in general, on John Carter, on the Avengers, some other movies, and also on virtual production which is a very interesting topic.

I will put together a small presentation for my co-workers in the next weeks - I made a LOT of notes but they would work better when organized and edited. I'll post a summary here for you guys too.

The short version is that everyone simulates a lot of stuff but there are some very interesting differences in the various studios' approaches. Everyone goes to Mova Contour (well maybe except for Weta), everyone uses blendshapes based on FACS as the main facial animation approach, although there are some differences in the implementations and almost everyone simulates fleshiness on top of the animation.

Also, Avengers is awesome and there's more full CG work in there than you'd think, including Mark Ruffalo in one shot.

Adam Baroody

11-05-2012, 03:06 PM

Tamas, unfortunately I use softimage so I cant make use of that script. I really wish softimage had artisan. I'm using mudbox for paint deformation (zbrush for initial sculpt).

Do you have an opinion on which style of edge flow you would use?

vargatom

11-05-2012, 08:13 PM

Well, the other thing you can try is a sort of indirect skinning. Softimage should have something like a wrap deformer where basically you use a mesh instead of a lattice to deform another mesh.

So you build two versions of Spidey, one with the seams modeled in, which you texture and sculpt and render in the end.
Then you remove the extra loops around the seams to smooth out the mesh. So you sort of keep the middle loop of the seam, but even that should be flattened out. You put all your rigging and corrective sculpting on this one and have it hidden in the scene. This mesh will drive the high res mesh using a wrap deformer.

We do this all the time for our simulated cloth, because seams, thickness and other model details would mess up the sim, so that needs a single sided flat surface too. The riggers do their work with that simplified geometry and then they wrap the highres textured version on top of it. Also makes the simulation faster so it's a double win.

Oh and I'd certainly model the seams into the model!

Rox

12-05-2012, 10:37 PM

Hey Tamas, could help with this?

How do you deal with tricky body parts like armpits, shoulders and such? I mean, sometimes a character moves his arm very close to the body and it interpenetrates with it. It's okay with very simple character, but when you have simulated clothes and moving armor elements everything literally explodes. Do you somehow simulate collisions on the body or use muscle collisions there or something?

vargatom

13-05-2012, 07:39 AM

I'm not that much into rigging apart from what we need to communicate about with the guys there, so I'm not 100% sure about what they're all doing. They're quite clever but we have very short production schedules so there really is no advanced system in place, they deal with everything on a case by case basis mostly.

We usually have a skinned naked torso in every character reusing the same base mesh, this is the collision mesh for cloth sims. I'm not sure if there's any muscle system involved, we certainly don't have the time and resources of big VFX houses like Weta or ILM to build detailed simulations. We also do some default pose sculpts for things like knees and elbows. I think the armpits are usually imploded a bit when the arms come down to leave enough space for the clothing.

There's also a lot of shot based fixes going on, like pulling the simulated cloth geometry around with clusters, putting animation controls on top of the shoulder armor plates and tweaking them until any intersections are not seen from the camera, and so on. Sometimes you can also get away with a little bending on solid objects, if there's enough character and camera movement.
The key here is that we don't produce any complete features, all the characters have very limited appearances and the amount of time required for a "perfect" rig is usually more then just fixing what looks wrong. So it's more efficient to pick your battles, compared to looking for ideal solutions.

Movie VFX would be a very different thing. Weta simulates every creature, from skeleton to muscle to flesh and fat, even in Tintin; and all the sims can also interact, so the floppy second chin of a fat character would automatically deform his shirt when he's talking.

Double Negative also did a lot of sims on John Carter but they chose to do simple and fast, and then fix hero shots by caching the deformations and adding sculpting to tweak the shapes and muscles and all. I guess the argument here was that they had a lot of shots with a LOT of tharks and they had to keep sims simple enough so that they can run all the jobs during the show. Then whatever looked wrong in the foreground could have been fixed pretty fast by a team of talented artists with a simple tool. It's one thing to simulate 10.000 hairs on Thor because he's alone in the shot, but a hundred Thors would had to deal with a lot more simplifications.

Or another example, the scene in X-men First Class where the Beast guy tries to cure himself and then he transforms into a hairy blue creature and there's a long POV shot of it. They were at about 70 iterations on the cloth sim before they decided to change a few things and try again. That level of refinement is impossible on game cinematics, and while expectations are getting higher and higher, we still don't have to deliver that quality yet ;) but we are constantly looking at more advanced solutions. We're growing, software gets better, machines get faster, so we work on improving our stuff all the time, and may be simulating everything in a few years, too...

vargatom

13-05-2012, 07:46 AM

Oh and I'm sure big studios do a lot of the little tricks too. Check out Iron Man 1 and 2 - the armor bends around a lot, stuff rotates out of the way magically, and if you can't actually see the whole piece then you're not bothered if it's like a feet up into Stark's stomach or something :)

Rox

13-05-2012, 09:00 AM

Thanks very much Tamas :) Very helpful reply.

j3st3r

15-05-2012, 11:09 PM

Actually I created a Softimage version of PaintDeform. However due to the limitation of weightmap, it's not functioning well. I could recheck my ICE tree, and send it to you Adam, for checking.

j3st3r

16-05-2012, 02:15 AM

For Softimage users, http://dl.dropbox.com/u/16694969/SmoothDelta.1.1.zip
It is similar in working to the DPK Paint Deform, but since Softimage has no Artisan functionality, I can't do an exact copy of that.

Usage

1. Select source geo
2. Deform|SmoothDelta
3. Model your changes
4. Paint smoothing with weightmap painting
5. Ice tree smooth delta has an itereation parameter

works pretty well, but as I mentioned, not as nice as the DPK_PaintDeform

Cheers

Szabolcs

vargatom

17-05-2012, 12:34 PM

So, here's a short summary of the FMX presentations I've attended, first a quick overview and then the character related stuff. I'll try to make it brief - I made a lot of notes but most of it isn't that interesting IMHO, the presentations weren't as technical as last year and there weren't too many practical tips to pick up either.

One big industry trend is virtual production. The definition isn't entirely clear yet - the Weta guys were the closest with their approach of 'using the best tools available to make films'. So it's not just about using mocap and previz in combination ;)
A lot of the presentations focused on live movie productions, using simulcam and proper planing and working on set. This is really high end stuff, but not related to 100% CG animation; got me interested in Real Steel on a purely technical level though. Oh and Casey Schatz from Giant Studio gets the best presenter of FMX award - imagine a hyperactive kid on coffee, I was laughing my *** off while learning a LOT of interesting stuff. I'll try to google his name to see if there's any videos on the net...

The case study on Tintin was mostly about tying the animatic stage into the actual production. Using the same FX tools with scaled down parameters to make sure it's possible even with a realistic style; using the same departments to produce animatic and final assets to create a better understanding and workflow. They also mentioned needing better lighting in previz because sometimes they didn't use good light conditions for enough shots as it couldn't be seen in the virtual camera.
All in all, previz with mocap and continous refinement is a good production approach but I guess most animation studios have been using it for a while anyway (we've been starting with mocap and adding cameras after for years, too).

Working in the VFX industry is dangerous.
If you're at Weta and look like Tintin, prepare to do a lot of actual wardrobe fitting (they made 300 sets of actual costumes), then pose for reference video, get wind blown onto you, get put underwater, then rolled in sand. Oh and Gino Acevedo will take a life cast of your head to scan it for skin displacement textures if he thinks it looks interesting enough.
If you're at DNeg, you'll be collared with equipment from a sex shop to check how a 4-handed martian would handle you; if you're an actor playing that martian, you might break your ankle while walking on stilts. If you're a VFX sup, the desert sandstorms may nearly get you blind, if you're a witness camera you'll be destroyed completely.

John Carter was big at the conference by the way. It seems that everyone really loved to work on the show, and Disney chose DNeg mainly because the production crew wanted to learn with the VFX studio instead of getting schooled by them. Noone really mentioned the box office flop or any marketing or director related controversies though. Couldn't decide if it was because of politics, or genuine admiration for Stanton...

vargatom

17-05-2012, 01:00 PM

CG characters seem to have reached a level where a lot of stuff is already standardized, there weren't any real breakthroughs or divergences. I'd sum the tech up as the following:
- simulations for skeleton, muscle and flesh layers, and of course cloth and hair
- blendshape based facial rigs combined with post-process flesh/fat simulation
- optical analysis based facial performance capture translating data for FACS action unit based animation
- Mova Contour's tech involved at some point usually
- 'creature FX' means any kind of 'plussing' using various Ncloth based simulations/deformations, tension map driven animated maps for displacements or pigmentation changes, etc.

Practically every VFX studio uses the above mentioned tools in various combinations. The only real differences are how much each ingredient matters in the whole.

DNeg for example captures FACS shapes for the actors using Mova's tech and uses it as reference and in the face capture solver.
MPC used the actual Mova data as the base for their CG doubles in the Potter 7 scene where 9 characters transform into Daniel Radcliffe (they're all fully replaced).
ILM used the Mova data as reference for Ruffalo/Hulk blendshapes.

Now imagine a scale where one end is Weta; they believe in simulating everything, with a fully physically correct approach. They'll go as far as getting chimpanzee cadavers from the zoo to scan for their muscle rigs. Every creature is simulated, even if it's a tiny ape or if it's a stylized human in Tintin. Lots and lots of research and custom code are required.
The bonus is that you can get the floppy second chin of a talking character to deform his shirt when he's talking. Or you can squish an ape's face against a window easily.

The other end I'd say is Double Negative. They build a relatively simple model and do quick and dirty muscle sims to get a starting point. Then they use shot based corrective sculpting to fix muscle deformations and create proper body contact; animate a separate set of muscle-based deformers to fire muscles and work the neck; and use a lot of other small trickery on top for jiggle and skin sliding and such. So the basic approach is to work for the camera instead of trying to do it correctly all the time.

ILM was somewhere in the middle, using both more complex muscle and flesh sims and some post-process custom deformers to get there. Didn't get enough info on MPC's approach but it looked similar.

The results are 100% realistic digital doubles. I've mentioned the transformation scene in Harry Potter; but there was also the question on Avengers about where to place the hand-off between the live plate and the CG character. The simple answer was to just go all CG...
The shot where Captain America jumps out of the jet to follow Thor and Iron Man was one example. Full digital double work from the start by Weta (I'm told the assets were handed over from ILM - no talk about how they converted stuff though).
Also, Hulk's second transformation started with a 100% CG Mark Ruffalo, no live element there. Most realistic CG human ever, hands down.

So, Hulk. ILM believed that they should build the CG Ruffalo as a starting point because once it looked realistic it was much easier to tweak the character to become the Hulk - it was easier to find out what did not work when they knew that the human version was 100% real (it was). This also gave them a starting point for the transformations because of the identical base meshes.
They did everything - lightstage scan, lifecast and scan, teeth and gums and fingerprints, mocap, optical face capture; anything you can name that's a mean to get life based data into the computer. Then they worked on it in zbrush and such even further.
They had multiple versions of the model with different resolutions. The first rough one was for the muscle sim, then they added details for flesh sims, then more for the vein and tendon stuff. The face was incredibly detailed for the blendshapes, too. Starting from Mova data but hand sculpted.
Hair was important - facial, nasial, ear, chest, arm, everything. Stubble has a subdermal element that's also visible because of skin transparency, I guess he meant paint some stuff into the color maps and don't rely on the fur/hair only.
They also painted sweat maps with droplets and such instead of a general wetness layer. Displacements were separated into three frequencies both in the maps and in the shader and were tweaked individually based on camera distance to get the best look. Again some tension maps to remove blood color (in this case green) from compressed skin. They also started the comp by dialing saturation down to nearly zero and getting the contrast right first, before they put the color back in.

There was also something even better than Hulk - Green Steve. Since words can't do him justice, I'll wait until the DVD release and hope that ILM somehow contributes some video to the extras. Was definitely the high point of the presentation.

Oh and the big lesson is, don't do characters in primary colors. All of them are a bitch to light - Red Iron Man, Green Hulk, Blue Navi, doesn't matter.

So, that's the short summary. Feel free to ask some questions, but as I've said there weren't too many actual details this time. Only thing I've seen was a little extra Jake avatar facial stuff in the Weta general character presentation. But any recording was strictly forbidden and I wasn't keen on getting thrown out (also had a fever for the first 2 days which sucked).
Yeah and Stuttgart is lovely, I'll try to go next year again, maybe we can meet up and have a beer (it's cheaper than a Coke!)

ZippZopp

17-05-2012, 04:53 PM

great write up, thanks for the information!

j3st3r

17-05-2012, 07:45 PM

Great review, thanks for sharing!

tonytrout

17-05-2012, 10:00 PM

great overview, thanks vargatom

Infinite

17-05-2012, 10:09 PM

I've been eagerly awaiting your reviews Tamas. Thanks for taking the time to share the information, really interesting.

Did you notice, has the Mova tech improved at all over the years? is the data any better. There site has hardly been updated since '07 and their examples are quite average.

Kel Solaar

17-05-2012, 10:27 PM

Thanks Tamas!

Very good write up :)

Confirming MPC is proceeding the way you thought it was, we actually also have a dedicated techAnim department that can fix stuff on caches whenever it goes wrong.

KS

Intervain

17-05-2012, 10:42 PM

thanks for taking the time to write these down:beerchug:... yeah I remember seeing pics of those poor wet Weta people ;)

vargatom

18-05-2012, 12:06 AM

You guys are welcome :)

Did you notice, has the Mova tech improved at all over the years? is the data any better. There site has hardly been updated since '07 and their examples are quite average.

Can't really tell, but it's always been quite good as far as I've seen.

The MPC guys even extracted some low frequency animated displacement maps from the highres Mova mesh, their final head model didn't seem to be that detailed though.

They also don't capture eyes and the lips themselves, so you still need some modelers to cover that part.
I actually had a discussion about this with one of my co-workers there - I thought it should be possible to put together a nested blendshape rig where one of the shapes is wrapped to the Mova mesh, and another shape has a full eyelid and lips rig. Then you can add another shape that has a full facerig and use animated weight maps to blend between the various heads. Yeah, sounds kinda complex ;)

Anyway, it seems that Mova is now a standard stage in all CG character pipelines where it's based on a real living human. Except for Weta.
I think it's because everyone tries to get as many references as possible, scanning and photographs and lifecasts and whatever, and this kind of capture is very good reference for sure.
Oh and I think Blur also used it for the Batman Arkham trailer they did.

Infinite

18-05-2012, 12:35 AM

You guys are welcome :)

Can't really tell, but it's always been quite good as far as I've seen.

The MPC guys even extracted some low frequency animated displacement maps from the highres Mova mesh, their final head model didn't seem to be that detailed though.

They also don't capture eyes and the lips themselves, so you still need some modelers to cover that part.
I actually had a discussion about this with one of my co-workers there - I thought it should be possible to put together a nested blendshape rig where one of the shapes is wrapped to the Mova mesh, and another shape has a full eyelid and lips rig. Then you can add another shape that has a full facerig and use animated weight maps to blend between the various heads. Yeah, sounds kinda complex ;)

Anyway, it seems that Mova is now a standard stage in all CG character pipelines where it's based on a real living human. Except for Weta.
I think it's because everyone tries to get as many references as possible, scanning and photographs and lifecasts and whatever, and this kind of capture is very good reference for sure.
Oh and I think Blur also used it for the Batman Arkham trailer they did.

Interesting ideas Tamas. I personally see the Phosphorescent approach as dated now, I think more detailed captures can be acquired using a more modern approach. So interested to see how they fair over the coming years. I have seen their data first hand streaming and re-projected to a quad mask, quite amazing really.

I like your idea about the rig setup. I know more studios like using the head mounted capture rigs, rather than a sitting performance but I honestly think you can use this sitting 4D captured FACS data to fine tune, tweak and drive the intent of emotion of the likeness on the actual rig in someway. I have yet to see a studio supply 4D FACS data of a full sitting performance, going from Neutral to Pose back to Neutral, Neutral to Pose back to Neutral and so on. I think all this could be recorded very quickly and efficiently. Tracked and exported.

It does seem though like MOVA's tracking is powerful, much like Dimensional Imaging's approach.

But what Weta did with Avatar is incredible. The bar is set for sure, even though the characters are alien, you can see the Actors resemblance, IMHO better than Buttons. When I first saw Avatar I couldn't figure out why Uhura was Blue!! also Sigourney was near 1:1 as Navii. Cool to see a younger albeit "alien" version of her.