Spidey Sense aka Retopology

[Nacy Nightfire]

It occurred to me having spent hours watching retopology instructional videos, obsessively reading about  general topology edge loop flow and redirection and staring for hours at wire frame images, I must have been a spider in a past life.  My fingers itch to move vertices around.

Anyone else love this stuff as much as I do, or do most folks see retopology as as a necessary evil in their quest for the perfect 3d mesh?

Note: I live for a great UV unwrap challenge, as well.  It puts me in my "happy place".

[Pamela Galli]

I choose B. I used to love love love decorating in SL, too, until I had to decorate dozens of model homes. 

The thrill is gone. 

[Nacy Nightfire]

Remember that book "Do What you Love and the Money Will Follow"?  The part the author doesn't share (I'm guessing here as I've can't remember actually reading it) is for most people all work eventually becomes tedious.  Once one turns a delightful hobby into a job one will, over time, grow mighty sick of that hobby.  I always felt a better version of that concept is  "Do What you Don't Absolutely Loathe - Do it Better then Anyone Else and the Money Will Follow (so in your Spare Time or Retirement Years you can do what you DO Love)", but that title is not likely to gain traction.  

 

 

 

[arton Rotaru]

4 hours ago, Nacy Nightfire said:

Anyone else love this stuff as much as I do, or do most folks see retopology as as a necessary evil in their quest for the perfect 3d mesh?

I also like retopology. Since I don't do characters and the like, I don't have to worry too much about topology and edgeflow. So the retopo part is a rather relaxing task for me. The model is done already, and I only have to draw some polygons over it. 😀

UV mapping is fun also. I cut seams, and pack UVs by hand still. Makes me happy too, if I can scale the whole lot up a little compared to an automatic packing.

What puts me in my happy place is baking normal maps though. If the map comes out flawless, it's pretty satisfying to me.

[Kyrah Abattoir]

I enjoy the design part a heck of a lot more, but retopo is just part of the process. There is still the satisfaction of a job well done.

[Nacy Nightfire]

I, also, tend to model static objects and not characters, however I purchased the Blenderella tutorial from the Blender website a few years back and I finally had a moment to watch it.  It's currently available for free on youtube.   I find the meticulous work where the artist, Angela Guenette, who works over the model, again and again, moving the vertices almost an imperceptible amount in order to get a perfect result, mesmerizing.  It certainly puts my well intended, but what I now see are essentially sloppy modeling efforts to shame, now that I understand in detail how a professional works. 

Many training videos pass quickly through the necessary steps in building up a model in order to keep the audience from glazing over, but the down side of editing is you can't get the sense of how much time is devoted to getting the job done.  There are many time-lapse sections in the Blenderella series, but they are easy to follow.  Although the process in some areas are sped up, nothing seems to be edited out.   I can only watch for about 15 minutes before I am "itching" to  jump back into Blender and move vertices.repeat.

 

[OptimoMaximo]

On 3/18/2019 at 7:06 PM, Nacy Nightfire said:

do most folks see retopology as as a necessary evil in their quest for the perfect 3d mesh?

Most folks in SL see this as an unnecessary hassle for the perfect 3D mesh sale, IMO.

Some time ago there was someone here on the forums claiming that the SL limits on vertex counts are ridiculous, because he had to spend so much time in ZBrush (!) finetuning the ZRemesher to output something that would be accepted by the uploader. Leaving aside the lack of understanding about these ZBrush tools and what they're intended for within the workflow, the average SL user thinks that highpoly = high quality, for which every modeled bit of detail they're able to keep on the mesh counts towards a "higher quality product", and retopology is seen as a "dumbing down my hard work"... totally neglecting details like ZRemesher's tendency to create spiral edgeloops (!!), then wondering why their rigged content deforms weird in some areas or why it takes that long to snap into place before it starts animating along with the avatar animation XD

So yes, I do love the retopology process and work, and the satisfaction it gives me when the end result deforms, moves and fits flawlessly.

 

[Nacy Nightfire]

1 hour ago, OptimoMaximo said:

"details like ZRemesher's tendency to create spiral edgeloops (!!),"

Optimo - 

I agree that beginners to 3d modeling for SL do not really understand the intended work flow of Zbrush otherwise obvious to anyone at an intermediate or advanced level.  The whole process is extremely daunting (to say nothing about how extremely daunting Zbrush is as a program for beginners) and those starting out  "can't know what they don't know" about good practices.  There are probably a few workflow tutorials out there on Youtube, but I don't remember actually looking for them when I was a complete beginner. I stumbled on this kind of info as I went along the learning curve so I made my share of stupid mistakes.  I simply didn't know there WAS a workflow. ( But, I give my earlier "beginner self"  an "A" for enthusiasm. 😊).

I  believe the spiral issue with Zremesher has been corrected somewhat in Zbrush 2019.  For anyone who might read this that is interested in this Zbrush feature, there are ways to influence/improve Zremesher results such as engaging symmetry, drawing suggestion lines with the Zremesher guide brush or adding automatic guides by creasing and/or creating polygroups then going to the Stroke menu>Curves>Curve Functions and select Borders, Polygroups, and/or Creased Edges (then press the Frame button)  to create curves to influence the way Zmesher works.  And there's second algorithm for Zremesher to try by holding down Alt while you activate Zremesher which is worth trying for comparison to see which gives a better result.  Even with the improvements to Zremesher,  it's probably always going to be necessary to review and tweak topology automagically created to fix  tricky areas that need very specific edge flow for animations used in SL. 

I often use ZRemesher early on to quickly create better flowing  base geometry in preparation for subdividing for high poly sculpting. I don't enjoy sculpting on the random geometry created with decimation as I have beginner sculpting skills  and everything goes "rubbery" on me using a decimated model. Working out  this early topology seems to "keep me in my lane" as I sculpt.  Once completed I'd likely hand retopolgize, simply because it's so much fun.

 

 

 

[OptimoMaximo]

On 3/20/2019 at 4:29 PM, Nacy Nightfire said:

I  believe the spiral issue with Zremesher has been corrected somewhat in Zbrush 2019.  For anyone who might read this that is interested in this Zbrush feature, there are ways to influence/improve Zremesher results such as engaging symmetry, drawing suggestion lines with the Zremesher guide brush or adding automatic guides by creasing and/or creating polygroups then going to the Stroke menu>Curves>Curve Functions and select Borders, Polygroups, and/or Creased Edges (then press the Frame button)  to create curves to influence the way Zmesher works.  And there's second algorithm for Zremesher to try by holding down Alt while you activate Zremesher which is worth trying for comparison to see which gives a better result. 

I've just upgraded to ZBrush 2019 and i still have to look at the new ZRemesher, by the looks of things it's a significant change since there's a "Legacy 2018" button next to it. So far i can just tell about what I've tried (2018 and older), and the ZRemesher Guides do their job up to some extent. Edgeloop Masked borders and slice brushes to make polygroups splits work nicely but it requires full loops everywhere, ideally perfectly straight. Even using the 3 tools in conjunction, so far, has led to a few spirals in my experience. Sure thing the result is NOT meant to be used as a retopo model, rather it's a mean to convert a Dynamesh into a workable mesh with topology and start the subdivision sculpting for the tiny details (after projecting the subdivided mesh onto the original Dynamesh to catch details that may have gone lost in the process). Might be worth using bits and pieces of it in the retopology though, with the due clean up of the troublesome areas.

On 3/20/2019 at 4:29 PM, Nacy Nightfire said:

 Even with the improvements to Zremesher,  it's probably always going to be necessary to review and tweak topology automagically created to fix  tricky areas that need very specific edge flow for animations used in SL.

That's what happens in real game and movie productions too. Don't think that SL is a special case of and in any form. The platform has a few limitations in place for various reasons, but i see a common misconception for which there is a huge difference gap in making models for SL or for another engine, aside from the tech limitations. For one, the idea that bento rigging is any different from the legacy skeleton rigging, or that fitmesh is any different from regular rigged meshes. Again, aside from tech specs that may differ because of the platform implementation of feature that require a different amount of data to handle, there is no difference between the two things (bento vs legacy skeleton / rigged vs fitted) and model making techniques do not differ, unless we look at polycounts (whereother engines can go way higher than SL if needed)

On 3/20/2019 at 4:29 PM, Nacy Nightfire said:

I often use ZRemesher early on to quickly create better flowing  base geometry in preparation for subdividing for high poly sculpting. I don't enjoy sculpting on the random geometry created with decimation as I have beginner sculpting skills  and everything goes "rubbery" on me using a decimated model. Working out  this early topology seems to "keep me in my lane" as I sculpt.  Once completed I'd likely hand retopolgize, simply because it's so much fun.

I'll give you here a bit of workflow expertise on how this is supposed to be handled:

1. freeform sculpting with dynamesh, up to general shape is achieved and the low frequency details are in place. Polish it to remove bumps.
2. increase dynamesh resolution for mid-fine details sculpting, when done...
3.  duplicate the tool and ZRemesh one of the copies, subdivide it an project it onto the Dynamesh left over from the duplication (this catches the details lost in the process)
4.  finalize your sculpting on the ZRemesh result with all the high frequency details
5.  Duplicate the sculpted Zremesh mesh and bring one copy back to initial subdivision. This way you keep the low-res displacement from the sculpting and you can reuse this to "steal" good topology body parts
6.  Duplicate the finalized sculpted ZRemesh and apply all the subdivisions to get a real high density mesh
7. run the Decimation master on this, the decimated mesh is the one to export as retopology base (it keeps shape, details and doesn't choke your app when importing it into Maya, Blender, etc)
8. export the unsubdivided Zremesh result from step 4
9.  import both meshes to your 3D app and check the ZRemesher result topology to identify issues, delete what doesn't work and keep the rest
10.  work the empty spaces left over from step 8
11.  bake maps from the decimated mesh OR send the retopoed model back to ZBrush and perform the baking in there, remember to flip Y when exporting the texture.

Decimation Master's result is not intended for sculpting or animation at all, that's why you get wonky results by doing so.

[Nacy Nightfire]

Thanks Optimo! You are a gifted teacher. I've learned a great deal from your past Blender classes at Builders Brewery. I've copied your recommended workflow for future reference. It's very helpful.

My 3d skills are limited and "hodge-podge" as I rely too heavily on free Youtube and Vimeo videos for training. I realize there's alot of misinformation and poor work flow habits I've no doubt adopted that I will have to identify and "un-learn." I work in 3d only as a hobby, so try to limit what I spent, but I can see where I should invest in some good training videos authored by professionals.

RE: "and you can reuse this to "steal" good topology body parts." I've finally modeled/retopologized a decent human ear with good flow and only 22 edge verts to connect to a head model. I plan to keep that one forever so I never have to go through that challenge again!!"

[animats]

Mesh reduction with retopology is the way to go, but many of the tools for it are brittle. Some don't like meshes with errors. At one time I was trying out various mesh reduction algorithms, in hope of finding a usable open source replacement for the one in the viewer's uploader. The ones that are free crash on flawed meshes, and the ones that are robust are expensive.

[Nacy Nightfire]

14 hours ago, animats said:

Mesh reduction with retopology is the way to go, but many of the tools for it are brittle. Some don't like meshes with errors. At one time I was trying out various mesh reduction algorithms, in hope of finding a usable open source replacement for the one in the viewer's uploader. The ones that are free crash on flawed meshes, and the ones that are robust are expensive.

I'm curious to know what you mean by "flawed" meshes.

Blender offers selection options which easily target non-manifold geometry, attached interior faces  floating interior faces*,  Ngons and loose vertices, to help you eliminate these issues as a final "clean-up" step in the modeling process.  Note that to specifically  highlight tris, quads, and ngonsn  select "Faces by Sides" then press F6 to bring up the menu that allows you to enter faces with  the number of edges you want selected.

I may be misunderstanding what you are posting here,  but  others might find this a useful to addition to their Blender workflow to clean up any meshes that might cause errors (along with the essential CTRL A in Object mode to lock in location, scale and rotation of an object.)

* For any internally floating faces you would need to hover over the model and press L to select connected faces then hide them to see if anything remains unselected in the interior space and delete that then unhide the model.  The "select interior faces" applies to connected geometry that is in the interior.

Edited March 24, 2019 by Nacy Nightfire
Correcting mis-information

[animats]

Re flawed meshes and retopology: Yes, there are tools in Blender to help clean up meshes. I was talking about mesh reduction in the viewer's uploader. Terrible though it is, it generates most of the low level of detail meshes in SL. Look at the SL world with the LOD factor set to 0. If you can see through something, it was probably uploaded with the built-in mesh reduction. Everyone who understands what it does realizes it's a terrible algorithm. But it's both fast and robust.

A few months back, I tried some open source mesh reduction algorithms.  They worked well on simple, perfect geometry and tended to crash on bad or marginal geometry. Some algorithms need a watertight mesh, so there's an unambiguous inside and outside. One didn't like too many edges coming together at a vertex, as with the poles of a mesh sphere. It's not that this  problem is unsolvable, it's that code from academia tends to skimp on robust handling of the hard cases. And it's really hard to find anything that works decently when you demand reduction to 12 to 20 triangles.

We really should have an algorithm in the uploader which makes halfway decent lower LOD models. Ones where the lowest LOD always covers the entire volume of the original model. At least we wouldn't have see-through houses. Or we should have viewer side impostors, as Beq Janus suggests.