Jump to content
Sign in to follow this  
tweetiepah

Question: Average Poly budget for mesh Avatar

Recommended Posts

I'm working on a full avatar here and i'm wondering what's my poly budget for each level of detail.  My head base mesh, including the neck, is 212 faces (424 tris).  I plan on uploading that as my lowest lod, and sucessive sculpted subdivisions of that base mesh as the higher levels of detail -- meaning the higest level of detail, subdivided twice, would be 3392 faces (6784 tris).  I plan on the full body avatar being around 15k faces (30k tris) at the highest level of detail.  I'm new to second life and most of the information i've found seems to be outdated.  Is my avatar reasonable, or do I need to lower my budget?

Share this post


Link to post
Share on other sites

I personally consider ~10k tris (Yes, tris, not quads) to be the ideal polycount for the highest LOD. That tends to be enough to get a good silhouette, and not be an impossible challenge to get all-quad.

For example, the newborn avatar I'm working on has less than 10k tris and looks like this:Capture.PNG

The matcap is a poor simulation of how smooth it will look when materials/textures are applied in game. The normal maps will give it back the detail I sculpted in. Basically, it'll look as good as a model with double it's poly count.

So to sum it all up, I shoot for 10k and am happy as long as I stay under 15k. 20k would be my max, and that would be for something with wings or four arms or something else crazy.

I hope someone else will chime in about the lower LODs, because I'm not too knowledgeable on that aspect.

  • Like 1

Share this post


Link to post
Share on other sites

Thanks for your reply.   The head, after one subdivision, is 1696 tris, and I feel it looks good enough (see below for a rough idea).  If i created the full avatar at this detail I would be well within your proposed budget (and like you, I do intend on including normal maps).

The problem is that the lowest level of detail (the base mesh) is too low for a medium LOD and I need something in between.  Unfortunately there is no such thing as half a subdivision. I'm also wary of trusting any automated poly reducing method.

Short of manual reduction by deleting edge loops and such (Blender is nice about this and usually preserves UVs and so forth), what are the suggested methods of accomplishing this that give good results (preserved UVs, no anomalies with rigging, etc)?  Does SL's automated method of reduction do a decent job?

PS: Your newborn looks great.

Screenshot from 2013-07-23 15:54:17.png

Share this post


Link to post
Share on other sites

I use a mix of automated and manual edge reduction. To get to my high LOD. Note that this comes before UV unwrapping.

Edit: It sounds like you're using a bottom up paradigm, where as I use top-down. That is, I start with the highest level of detail and decimate the model for each lower LOD. I see no reason why your way shouldn't work, but personally I would find it slower and less intuitive. Blender can do a lot of the reductions for you. Also, as you pointed out, there is no half-subdivision.

For lower LODs, I would probably mainly stick to automated. If you use the decimate modifier on "collapse", it will preserve your UV seams and maintain volume well enough that you won't notice at a distance.

For the high LOD, I start by applying the subdivsions to a copied mesh (I always keep a backup with all the modifiers in tact). I sculpted up to a multires level of 4 in this case, which came out to around 1 million polygons.

Then I use the decimate modifier on Un-Subdivide. It is only useful on multiples of 2 -- you'll see why as you play with it. As you can see, after 6 iterations, I dropped to a workable 14k. Note that you could go down another 2 iterations and not really have to do any manual reduction, but I found that the automated unsubdiv caused too much of a loss of volume and detail. You want to keep your game-ready mesh as similar as possible to the volume of the high poly model for best results on your normal map. Also, Blender doesn't know where to leave more edge loops to allow for good deformations.

Capture4.PNG

To manually reduce, I look for main edge flows (loops that terminate at polls circled in red), select the next continuous edge loop and then pick roughly every other loop. Then just hit x and delete edge loops.

Capture7.PNG

What I can't illustrate in screenshots is how I choose edge loops to keep for the sake of deformation and holding volume. Always start with the easiest way:

Start by trying to quickly remove every other edge loop from different parts of the body. Note any areas that lose a significant amount of volume or shape. After that, see if you can make up for the loss of volume by moving around nearby edge loops with edge slide and alt+s to scale along normals. If you can't, undo and leave an extra edge loop or two there. You should almost always leave some extra loops around the pelvis, shoulders, mouth and eyes, off the top of my head.

It takes some extra TLC, but it's worth the extra effort for the most visible LODs.

  • Like 1

Share this post


Link to post
Share on other sites

I agree with about everything Rahkis Said, 10k Should be more than plenty to design an avatar with and remember you can substitute a lot of detail now with the normal and specular mapping. Something to consider when you look at your final design and crave for just a little more detail, its a way to add it without increasing the polycount.

Share this post


Link to post
Share on other sites

Edit: It sounds like you're using a bottom up paradigm, where as I use top-down. That is, I start with the highest level of detail and decimate the model for each lower LOD. I see no reason why your way shouldn't work, but personally I would find it slower and less intuitive. Blender can do a lot of the reductions for you. Also, as you pointed out, there is no half-subdivision.

Sort of.  I was planning on going in both directions actually, as multires lets you do that.  I was planning on going high to low for the sculpted detail.   I was planning on using either the base mesh or a mid-res mesh as a basis for baking the normal map.  Where I wanted to go from low to high was with the UV maps and the weight painting, as it's a lot easier to do that with fewer vertices.

I do thank you very much for your suggestions.  I'll experiment with "decimate" but will probably end up removing loops and sliding as you suggest until I hit my target.

I do have a question about your model, if you don't mind.  Why did leave the belly button on your game mesh rather than have a flat stomach and the belly button in a normal map?  Is there a reason I should do this on my model as well for other similar details?  Is this for clients who have normal mapping turned off?

Share this post


Link to post
Share on other sites


tweetiepah wrote:

I do have a question about your model, if you don't mind.  Why did leave the belly button on your game mesh rather than have a flat stomach and the belly button in a normal map?  Is there a reason I should do this on my model as well for other similar details?  Is this for clients who have normal mapping turned off?

Unlike adults (where I'd probably do as you suggested) newborn babies have bulbous belly buttons which are rather visible from the side. Not having it would detract from the silhouette, which can't be faked by a normal map.

 

In this case, that extra detail has a noticeable effect on the look and feel of the model, while having a minimal impact on the poly count. That's the kind of tradeoff you're okay making on the high LOD.

 

This is also the first place I will start reducing the polycount for the lower LODs, since it wouldn't be noticeable from a distance.

 

On Multires:

 

Keep in mind that the multires system in Blender has it's quirks(bugs).

The rules of thumb are:

  1. Only up-res when necessary to get the next level of detail.
  2. Never go back and alter a previous level. Going back and playing with earlier levels will often cause artifacts in your higher levels that cannot be fixed without ruining all of your details.

Blender is getting better all the time and it's possible you won't run into any problems, but if you do, sometimes "apply to base" (alters your base mesh to better match the highest subdivision level) in the multires modifier panel can save you from artifacts. They usually occur in areas where the multires mesh is displaced too far from the base-mesh. Just remember that your base mesh as it originally was will be lost forever unless you saved it first.

 

It sounds like you're on the right path. Just be prepared to roll with the punches; making an avatar is never easy and there are many setbacks on the horizon. If you're ever stuck, this is a good place to ask questions.

  • Like 1

Share this post


Link to post
Share on other sites

Unlike adults (where I'd probably do as you suggested) newborn babies have bulbous belly buttons which are rather visible from the side.

Interesting. I never knew that.

Never go back and alter a previous level. Going back and playing with earlier levels will often cause artifacts in your higher levels that cannot be fixed without ruining all of your details.

I haven't used multires since Nick Bishop first first added the feature (years ago).  Since then i've been on somewhat of a break from 3d until me husband mentioned I could make some extra cash making and selling avatars and such to furries.  I used to know Blender (and much of Maya) backwards and forwards but when you don't use a skill after a few years, stuff changes and you get lost.  Getting used to the 2.5 interface change has been unsettling to say the least.  I'm sure it'll grow on me, but right now, i'm still used to 2.4x.

As such, I am somewhat rusty and really really appreciate your input.   I understand since the time I last used multires that Nick Bishop has mofified the codebase of multires so that displacements are now done relative to the normal in tangent space, rather than absolute XYZ displacements that break when deformed.  At least before, I was under the impression modifications could be made non-destrictively to lower levels of the subdivisision so long as the topology doesn't change.  Is that no longer the case?  If so, i'll avoid using multires and just subdivide manually and un-subdivide as you suggested.

Share this post


Link to post
Share on other sites

All I really meant to get across was that you should be careful as you work and be ready for setbacks.

My workflow involves sculpting in almost every step of the modeling process. I don't know how sculpting would differ from subdiv modeling on a multires surface...You don't necessarily need to abandon multires.

There is a whole can of worms here, so I'll try to keep it brief:

The entire cg industry moved away from vert pushing to a sculpting based workflow all in the course of like one year -- probably starting around when you stopped a few years back. There is a reason for that -- the old way was slow and unnatural. I suggest you throw everything you know out the window, download 2.68 and start catching up with the times (CGCookie is a good place to start). In the end, your work will greatly benefit from doing so. You don't need a tablet to sculpt.

I would also reconsider starting out with a commercial full mesh avatar, furry or otherwise. There is a reason why there are so few on the market and it's not due to lack of talent. It's due to lack of interest.

If you really want to make mesh avatars, I suggest making mesh heads, hands and feet and just create skins and shapes for the default sl body. That way it's not all just one mesh object and you have to take it or leave it as it is.

Just my 2 cents. I don't want it to sound like I'm telling you not to do what you want. I reserve the right to be completely wrong :)

  • Like 1

Share this post


Link to post
Share on other sites

Since I know NOTHING about making a mesh avatar *wink* I wanted to comment on the comments relating to normal and specular mapping for details.

While I am using those to some extent in my work, I think it is EXTREMELY important to realize that it will be a very long time before the average "joe" will be seeing the new materials. Firestorm with a humongous piece of the viewer pie is in no hurry to add it into the mix because of current issues (this from their devs on a blog that I read recently). That means, for me, that I have to log into the Linden viewer in order to add materials. I do not see them normally.

Then you have to consider that a large portion of folks using the official viewer (and again that isn't a big percentage) don't have machines that will run materials or simply don't know or care.

I am very excited about materials, but to base your work on them IMHO is a very bad plan. Too many people will never see what you are seeing  -- some will down the road. And if someone buys an avatar thinking it looks great and their friends take snapshots of them using a non-material setting ---- ?? Well you get the idea.


So plan for the future but I don't think it is the time to put all you eggs in the materials basket. Make sure your avatar looks great to those NOT using them :D.

 

  • Like 1

Share this post


Link to post
Share on other sites

I agree. All good points.

Your process need not change, however.

Use only enough mesh density to maintain a good silhouette all around and good deformation in the highest LOD. Make up for the lack of polygons with a high quality diffuse texture.

Follow those "simple" steps and those who can't see your normal maps will still see a good model. Consider this: 10k tris is still more than the default avatar has and with the right textures the default avatar can look pretty incredible (and it cannot use normal maps!). I wouldn't suggest avoiding materials just because some people wont see it. Even if it is a feature most of your customers won't use for a while, it's still a feature that will drive more sales over time.

Again, I definitely agree that it would be silly to make something that looks like crap without materials!

  • Like 1

Share this post


Link to post
Share on other sites


Rahkis Andel wrote:

The entire cg industry moved away from vert pushing to a sculpting based workflow all in the course of like one year -- probably starting around when you stopped a few years back. There is a reason for that -- the old way was slow and unnatural. I suggest you throw everything you know out the window, download 2.68 and start catching up with the times (CGCookie is a good place to start). In the end, your work will greatly benefit from doing so. You don't need a tablet to sculpt.

I'm totally comfortable with a sculpting workflow and have a tablet (though i'm probably rusty).  It's just I'm used to getting my topology and edge loops in place (usually I sketch a rough mesh on photos of a subject), and then use that as a basis to sculpt.

With my avatar here, i just haven't gotten to the sculpting stage.  I realize that I could work with something less than optimal and then retopo that once it's sculpted, but then I'd likely have to build two meshes instead of one (one to sculpt on, one for the retopo).  It's just the way i'm used to.

That being said, I used to have a lot of fun with 3d coat's voxel sculpting where you didn't have to worry about topology at all.  The app also had some really well done retopo tools and the normal map baking was at the time better than Blender.  Like blender, however, I haven't used it in years so I couldn't tell you how good it is anymore.

In the future I may very well use a pure high to low workflow.  The new Dynamic Topology sculpting in Blender seems very cool.  Once it matures some more I look forward to using it but right now i'm not sure how I feel about the "detail size" being defined in screen space.  I'd rather it use world/object space so no matter what zoom i'm at, it's creating the same level of detail.  That's closer to how 3d coat's voxels worked, although it's not entirely comparable.  RIght now it seems it would be very easy to accidentally zoom out, make a modification, and erase any carefully sculpted high frequency detail.  Have you used this feature?

 


Rahkis Andel wrote:

I would also reconsider starting out with a commercial full mesh avatar, furry or otherwise. There is a reason why there are so few on the market and it's not due to lack of talent. It's due to lack of interest.

Interesting.  What would be the most popular/profitable types of items in second life, then?  I'll probably continue making a full mesh avatar just because I like a challenge, but if that's not where the money is, i'll definately keep in mind other stuff in the future.

Ps: I don't at all take offense from your suggestions.  Even if we disagree on workflow or whatever, I still appreciate your input and you've been very helpful.

Share this post


Link to post
Share on other sites

I wouldn't hold my breath for Dynotopo to "mature". It's pretty much as feature complete as it is going to be for the foreseeable future. If I recall correctly, the main developer has moved on to a real job!

Despite that, dynotopo is my new favorite thing about Blender, hands-down.

The zoom issue doesn't bug me much, just don't use it for high-frequency detail. That's not really what it's meant for and the viewport performance starts to get iffy real fast once you get up in the million poly range.

My workflow is: Dynotopo concept, retopologize, add Multires and shrinkwrap, sculpt high-res on multires mesh, down-res for game-ready model and Bake normals.

Here's a good thread that contains over 100 pages of people testing out what is possible with this "immature" feature. :)

As for what would be more lucrative in Second Life, the last person I'd turn to advice from is myself, to be honest.

Avatars are definitely a good market to get into since there will never really be too many people talented enough to make good ones. Plus there will probably be no end to the untapped niches in that market.

Just keep in mind the limitations of mesh. All the people with furry avatars I've ever met (including my girlfriend) have heavily modded theirs to the point where it could be made up of 2 or 3 different avatars and you wouldn't have any idea where any one part came from. If you market a full mesh avatar that is all mesh in one piece, there isn't much you can do to mod that. Heck, there isn't much you can do to clothe that, for that matter.

I think it's going to be pretty common for the head, hands feet and other extremities to be mesh but to just provide an alpha, skin and shape for the default SL avatar's body. That way people can still mix and match to their contentment.

If you want to also provide an optional mesh body on top of that, that's just an extra perk. It's up to you to decide whether that extra effort will be worth it.

So, yeah. I didn't mean that you shouldn't make an avatar, just that you should think twice about it being fully mesh all in one piece.

  • Like 1

Share this post


Link to post
Share on other sites


Rahkis Andel wrote:

So, yeah. I didn't mean that you shouldn't make an avatar, just that you should think twice about it being fully mesh all in one piece.

I gotcha.  In that case I just might just chop the avatar up and sell the pieces.  Should I budget differently for this, or just make my full charachter ~10k and let the chips fall where they may when it's all chopped up?  I personally tend to use most of my budget on the top portion of a charachter (head, arms, chest, etc...).  Should I change this, and if so, what should be optimal budget for a head?

It's a shame if there are not likely goign to be no more improcements to Dyntopo, but your workflow does sound like a good one. I'll give it a shot next time.  It's similar to what I used to do in 3dCoat, although I usually put the HF details on the voxel model before retopo and baking.

Here I am at mid-sculpt (actually more like just started):

Screenshot from 2013-07-24 16:45:58.png

I also haven't yet purchased Avastar. What's your opinion on that?  Is it worth it, or will I be able to do the same things without it?  I'm particulary interested in using attachement points as extra bones.  Is that a nightmare without Avastar?

Share this post


Link to post
Share on other sites

Chopping it up shouldn't affect the poly count for any of the pieces, except in that you will need some extra polygons to fill in the holes. Some parts will naturally be more dense than others because they need more loops to deform gracefully.

Unless I'm thinking of someone else, Nicolas Bishop seems to have been the only one working on Dynotopo and so it would be hard for another developer to just pick it up anytime forseeable. That said, I'm sure it's not done evolving for good.

Anyway, it's as good as it "needs" to be for my usage. At least for now.

And yeah, I highly reccomend Avastar. I don't have it personally (yet), and you could surely make do without it, but I think you'd make your money back pretty fast in the time it could save you.

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Sign in to follow this  

×
×
  • Create New...