Direct Subdivision Manipulation (DSM?)

[TheIvIaxx]

link is broken

[ambient-whisper]

its a bad idea in my opinion.
the cage is way too spiky and if you want to do some good facial animation with it, it will break. more detail to the head is neccessary for proper skin sliding and such.

DSM is mainly good for final touchups. but to work with it too much is a bad idea because you never know where the verts will go on the REAL cage. rigging becomes harder if you got too many spikes too.

[PetterSundnes]

Which link is that TheIvIaxx ? The one to the .OBJ file or the one to the Andrew Loomis books? They both work now it seems. By the way, the .OBJ file is text based, so if you just click on it, the contents of it will be shown inside the browser.

ambient-whisper, think of the vertecies of the un-subdivided polygons as the control points of a nurbs curve, instead of the actual polygons you are modeling on. It's not the un-subdivided polygons that matters but the ones that are subdivided (the polygons which is going to be rendered).

If you model with a DSM style (directly tweaking the curves of the subdivided model), instead of the un-subdivided polygons, then the "real" polygons will be the ones that are subdivided (the ones you are working with). This is the esscence of what I am trying to convey here. If you have an application where you can directly tweak the iso-lines/sub-d-edges/sub-d-points without searching the bounding polycage for the right vertex, try it and see how natural it feels. Its a rather new feature in most apps, like C4D and XSI, but LW have had it for some years already. Silo and modo have this ability too (Maya too?).

I'd compare it with clay modeling with either a ten foot pole, or digging your hands right in there.

When the overall flow and proportions are in place, details like wrinkles and muscleflow can be further refined by adding more edge-loops and splits.

Subdivision of model with 204 polygons (LightWave MetaNurbs):

http://psinteraktiv.no/pic/face-morph.gif

http://psinteraktiv.no/pic/subdivision-anim.gif

[ambient-whisper]

yes i know which ones are to be rendered. but the idea is still really bad. :) because the REAL cage is on the left and its way too spiky. good luck making GOOD expressions where skin slides over bones... etc. right now there is way too little information on the model to do much with it.
your morph targets will not like what you do because the original verticies will be flying all over the place when you mix the targets. or even if you rig the face with bones and what not. you will need to make sure that your rig is extra flexible to allow the spikes that you have in there at the moment.

i agree that the loops are very nicely laid out but theres just not enough of them to do too many useful things with it.

trust me. dont let the technology fool you into thinking its a great idea ;). i used to think its good too, until i started to export models from one application to another. texturing, rigging, facial work, etc become a pain because i need to get my models into a workable stage first.

theres a good reason why computers are becomming more powerful. one of the reasons is to be able to display more polys on screen.. not less :)

gollum on the other hand has too many polygons in my opinion ( bay even said so himself ), but you probably remember how expressive the character was. the way the skil slid accross the cheeks. the way the lips puckered. the way he was able to hold his breath. etc..etc.

http://ambient-whisper.cgcommunity.com/images_dump/neato/gollum-7.jpg

also with more polygons you can add subtle variations to his skin as well as do better asymmetry. where if you pull one vertex it will change a small thing and not affect a wider proportion to make your character look like an orc :).

making clean cages like this ( where the polygons are so nicely laid out ) makes working in zbrush much much easier as well.http://209.132.69.82/uploaded_from_zbc/200407/user_image-1089870557tkd.gif

turns into...
http://209.132.69.82/uploaded_from_zbc/200407/user_image-1089746374ezr.jpg

if you do use these subdivision types.. use them only for previewing/ and tweaking the final stages of a model. but try to make sure that the model has enough polygons on its own that it wont lose too much volume when its subdivided, and it could stand up as a solid model even without subdividing.

cheers.

[PetterSundnes]

Most beginners in 3D modeling are starting off with too much detail straight away. The flow of the polygons and proportions are all off due to the focus on details. Its just like classical drawing, start with the flow, and then add detail to it gradually. At what level you choose to stop detailing a mesh is up to you.

but the idea is still really bad. :) because the REAL cage is on the left and its way too spiky.The real cage as I see it, is the final model, not what you start out with. The "spikes" arent there, when subdivision has taken place: http://symbolcraft.com/graphics/subdivision/
Just think about each polygon in your model, subdivided or not, are all spiky. Its when you apply smoothing of some sort they become a flowing surface (subdivision and smooth-shading, normal-maps ++).

where if you pull one vertex it will change a small thing and not affect a wider proportion to make your character look like an orc :).That is exactly my point here. One should affect the proportion and flow at an early stage, just when they are the most important to get right (as with a drawing). When you have a good flow through the model, then you can refine the model and tweak the details.

I am not saying you are wrong (not that I can either, considering your skill level :)), or that one way is supreme to the other, I am just trying to show people how they dont have to start off with a complex polycage and moving a tonne of vertecies around, but start off by getting the flow right. Then "freeze"/refine/whatever-you-might-want-to-call-it the model and continue on the finer details. The ability to access the vertecies on its subdivided level in the early stage is very valuable, instead of searching for them in the spiky "REAL" cage as you call it :)

Too many times have I seen people (as well as myself) start off with a box with a certain amount of segments in all directions, and then try to find the right position for each of them. The result is quite often rather boxy looking or just a messed up flow in general.

How about this for a rule: "A good polyflow with a right amount of details at the right places is all that is needed to create a good model, regardless of technique used." :)

This is a continuation of the lower poly mesh:
http://petter.ms/forum/head-c4d-render.jpg

[thedaemon]

The main problem with modeling like that, is when you start adding in more details. When you add a few edge loops, your model now looks funky because of the over compensating 'spikey" polys from before, so you have to bring them back now to compensate for the new defined edges. Its a waste of time (money) to model like that if you are going to add more detail in.

[io,]

i have to agree with thedaemon and ambient-whisper. and i know that is not that easy to rearrange the new loops that u just added. also, if u somehow decide that u want to use some obj as influence, it's not recommended to have a 10 cv character. it will not work. beside that, when u have a nice topology, rich enough to satisfy u'r shape needs, u have to think if u what that model to be deformable and animateble, witch might be deferent. imo, a good topology is a topology that is not to heavy to take ages to build an expression, but enough to be able to do whatever i need and to be predictable in deformation.



Dada

[eek]

Ambient-whisper's correct,

Its far easier to added loops and more detail to get the desired facial shape/animation working. This is what ive found over the last two years on my rig, and im working with unsubdiv models,rigging them then subdiving them. Its also highly essential to keep volume
( a major headache with gollum), and the edge loops must be accurate to accomodated muscles movement, and combinations of muscles.

good luck making GOOD expressions where skin slides over bones... etc. right now there is way too little information on the model to do much with it.

exactly!, skin compresses, bulges, and does all manner of things. With the key being that none of it intersects. Bone to muscle deformation and skin to fat deformation are highly complex, with and added layer of complexity being that muscles have a hierachal order, where muscles take president over others in there action. I.e they trigger different muscles in key orders.

For this to work you need a highly affect poly model and good edge loops, look at gollum: it had 10,000 individual muscle shapes, with 900 combinations of these.

eek

[Julez4001]

Eek! did you work on Gollum?

[eek]

No, sadly not.( i wish i had!) Ive just done a lot of research into gollum, FACs, Elkmans research, edge topology, SubD surfacing, hierachal muscle orders, fat and skin deformation, surface volume etc, etc etc.

I would of loved to work on Gollum, heck i would love to work on King Kong!

eek

[ambient-whisper]

I would of loved to work on Gollum, heck i would love to work on King Kong!

eek
what can i say. apply! ;)

[gga]

The main problem with modeling like that, is when you start adding in more details. When you add a few edge loops, your model now looks funky because of the over compensating 'spikey" polys from before, so you have to bring them back now to compensate for the new defined edges. Its a waste of time (money) to model like that if you are going to add more detail in.Well, it really doesn't matter if you develop a proper edge loop tool.
If anyone with some good ARexx experience can locate and decode (as it was never released in source but in bytecode form) my Lightwave macro collection for the Amiga from many, many years back, you will find an algorithm for doing loop splits with autocompensation, which, much to my surprise, has not yet been re-done in any modeling package yet.
Albeit the algorithm was pretty much ad-hoc (didn't know much about how subds worked back then), it did function relatively well as the equivalent of a nurbs isoparm insertion, automatically compensating when adding edge loops, thus easily allowing you to keep the previous shape of the object.

[optical op]

Agree that sub D is not the best method for creating a model which will later be used in animation, but if you wish to knock up a quick head for example that is simple gonna be used as an inanimate object, why not? I think that was really the point of the original post. You could even use this method to knock up a quick pre-vis model to use as a template before you go and make your beautiful, poly-perfect one which will be used in animation. Many Max users I know are very jealous about not having this type of modelling in their app.

Ta

[PetterSundnes]

No one is stopping anyone from freezing their model at any given time. My point is that it is easier to create the overall shape/flow with a few polygons to start with to get the basics right, then continue from there (wether its frozen or not at that point, is up to the artist).

Too many times have I seen people struggle with their 3D modeling due to overly high polycount. Keep it simple for as long as possible is my advice.

As a sidenote, both modo and Silo are based on working directly on the sub-d iso-lines, and I do not think it is for a reason such as: "because its cool".

See attachment for sub-d to frozen state (real polycage as its been called, what you see is what you have :) ).

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.