The big Blender Sculpt Mode thread (Part 1)

[Metin_Seven]

@m9105826

Yes, the code is really in that bad shape. The entirety of Blender could use a complete redesign and rearchitecture and would benefit greatly from it. Cycles is by far the cleanest bit, and it’s not even particularly clean. Alas, the manpower simply doesn’t exist to do the job that needs to be done. I fear that the technical debt in the code base taken on every time a new feature is added will eventually be the downfall of Blender.

This is exactly why I think it would be sensible to slow down with adding new features all over the place, and focus more on improving / rewriting dated but valuable parts of Blender.

Of course introducing new features and receiving praise from the community is much more exciting and rewarding than fixing stuff under the hood, but the Dyntopo case is a fine example of what happens if you add layers of new stuff upon non-maintained old components: eventually features that are used and valued by lots of users get deprecated, and not always replaced by something better.

In ZBrush just about every function that has ever been added is still functioning and keeps being usable in conjunction with newly added features, which enlarges the creative possibilities.

[dan2 (dan)]

@Metin_Seven

it would be sensible to slow down with adding new features all over the place

Exactly this. Sculpt mode has improved miles compared to 2.79 but also heaps of things broke and many features where never there in the first place. It’s great to get attention with shiny new brushes, but if people from other packages come over to blender to try the cloth brush just to find the whole multires in shambles it does more harm than good. They wont come back for another year at least.

They should just do some maintaining, cleanup and rewrites, I don’t mind how long it takes, but if they have a good foundation to build on afterwards it’d be great for everyone.

[avpiedra (Arthur Valverde Piedra)]

Hey all!
I figured out this will be a good place to ask for ideas. I am on quick and dirty concept side of modeling and sculpting things. 3Dcoat is particularly good at doing mesh (voxel) extrusion with stencils, so I thought I will try to replicate it inside Blender. I think the hardest part is speed (but it’s managable) and lack of square brush or rectangle select tool that would act as brush. With all the magic that team and Pablo Dobarro pulled off I think there’s nothing impossible, but I am also aware it’s not a prioroty.

On the sidenote, the Mesh Filter tool is really handy I just wish it could have a texture/stencil capability (yes it can be used with mask).

Here are speed up results. https://www.youtube.com/watch?v=XVqGqykrUA4

Also some shots:

Have a look and tell me what you guys think, please.
Also take a note I am a noob when it comes to sculpting Cheers!

[zeauro]

@Metin_Seven

Dyntopo case is a fine example of what happens if you add layers of new stuff upon non-maintained old components:

Not really. Dyntopo was released in 2.66. Bmesh was added in 2.63. That was new.
Bmesh was certainly not non-maintained or old at that moment.
Sculpt mode was not orphaned, either at that period. There was a developer : @Psy-Fi

Dyntopo was working at that moment. Last addition of sculpt mode in 2.80 before Pablo’s arrival was 2D FallOFF, Topology Rake and Manual Detail method for Dyntopo.
Two feature are a Dyntopo only feature. The other one (today renamed Projected FallOff) releases its full potential when used with Dyntopo like Snake Hook brush.

Pablo added lots of features that improved experience with Dyntopo.
So, Dyntopo has never been as good from user perspective.
Code is probably complex, undocumented and for that reason, hard to understand.
But it works. I have difficulties to imagine that what is delivering in Blender is completely messed up.
People are also saying that Blender’s Interface is a mess when they don’t take time to learn it.
And for a part of it, it is not false because it is WIP.
When a module becomes orphaned like Sculpt mode, there is a chance that some part of code stays as a WIP that nobody knows how it was supposed to be achieved.

Pablo is saying that he has to wait Campbell to fix things for him. I don’t know if he is complaining about review process. Or if he has a lack of knowledge about Dyntopo code.
But by taking responsibility of Sculpt Mode, he took responsibility of Dynamic Topology, too.
If Deprecation happens, it would simply be declared because of a lack of will to respect their own rules or to organize themselves to transmit knowledge from one dev to another.

Deprecating Dyntopo is clearly not what users, who followed its progression, were expecting.

Idea of Dyntopo comes from Unlimited Clay which was inspired by Sculptris, that was a fresh freeware at that moment.
So, Dyntopo is more an example of what happens when you enter into the unknown, you explore a new way of doing things and try to figure out how to push limits further.
When it was introduced, there was no idea that should stick to concept, basemesh sculpting.
It is a limitation that was installed in people’s mind by experience.
It is hard to sculpt a mesh with millions of polys with dyntopo. That is all.
Was the goal of dyntopo to reach millions of polys ? No. The goal was to explore new ways of improving sculpting experience.

In fact, volumetric sculpting development will probably be similar to Dyntopo development.
It will encounter difficulties and when limits will be reached, users will say that is limited to a certain use.
Dyntopo is not old technology full of hacks. It did not last more than 20 years like Blender Internal.
Dyntopo has less than 10 years.
Probably, there are things to do to clean it, document it and make it easier to maintain during again a decade more.
Blender still has metaballs. When somebody complains about metaballs, devs are fixing them.
If metaballs are not deprecated, it is difficult to accept that for Dyntopo.

[zeauro]

@avpiedra

lack of square brush or rectangle select tool that would act as brush

Pablo introduced a Tip Roundness slider in master to make tip of brush squared. That will be available in 2.83. You can use a texture in previous releases.

Maybe you are thinking about something more similar to Dots of Anchored Stroke method.

[lsscpp]

If I remember correctly Pablo wrote something about an abstraction layer between brushes and the geometry that makes easier to design and code sculpt features.
Maybe in the future we’ll see some decimate option on some brushes?

[avpiedra (Arthur Valverde Piedra)]

Hey thanks for the reply! I think I will record video on how it works in 3dcoat and write longer post with examples to illustrate the idea. Thanks for the tip on brush roundness, it doesnt solve my problem but it’s cool feature.

For now you can see how Eytan Zana uses stencils to extrude form from geo. Starting from 00:26 https://youtu.be/vgSWMGoCCn4?t=26

[xan2622]

Sculpt Mode Features Update — Developer Blog

[CLuss]

What about the polycount subject?
Don’t wanna sound like a broken record but I rarely see/read somebody mentioning it and is a sad surprise for me.
Thanks

[TheRedWaxPolice (I know dead mesh when I see it..)]

@CLuss

What about the polycount subject?

That’s a tough business…

[zeauro]

@avpiedra

For now you can see how Eytan Zana uses stencils to extrude form from geo.

Thanks for clarification. Indeed, that looks very efficient.
Blender can be close but it requires many more steps.

You can use a Stencil mapping for a brush texture. Then, you can keep texture overlay visible by keeping brush as active tool and call Box Mask by pressing B.
Then, you can invert mask and use a big radius and a Dot stroke method or Anchored method to obtain a uniform result.

You are perfectly right. Adding a modifier key to mesh filter tool to do a Box Filter or Lasso Filter with supports of textures would be a lot quicker.
Mesh Filters are perfectible. I suppose that several things that Pablo added are.
For instance, Box Mask active tool simply does not New, Extend, Subtract, Invert, Intersect modes like Box Select tool. But it would be pertinent.

[Rodger_D]

He has shown a glimpse of the brush system using animation nodes. Animation nodes can do magical things for sure. So, who knows to be honest.

This the first prototype of the Node Brush. It runs a node network to calculate the displacement on each vertex so it can produce all kinds of deformations with unlimited parameters. Based on a WIP branch by @JacquesLucke #b3d pic.twitter.com/JHe9PNGkXV

— Pablo Dobarro (@pablodp606) February 13, 2019

`

[CLuss]

This should be high…

but better than nothing.
Thanks

[Rodger_D]

For the time being. It shouldn’t. There are more performance issues that needs to be tackled first.

Starting at 28:54

---

---

Performance in general 27:16 - 32:30

[polynut]

I know but for fingers it is a working solution until we get something better.

[Ace_Dragon]

For high quality sculpts and especially creatures, the idea of being able to get a somewhat detailed shape down with Voxel remeshing and following that up with a fully functional multires session is critical to have. Performance will only get you so far if everything you make caves in from bugs.

Because of the fact we have the corrective smooth modifier, even a basic rigging of the thing would be quite viable, it would be even more so if the new multires does not lock the mesh in place.

[Nur_Rp (Rahman)]

when it comes tiny detail dyntopo is realy weird, you cant even blockout properly.Tiny details matter
dyntopo (verts: 98K)

while woxel remesh does the job okish
(verts:39K)

[julperado]

Did you use dyntopo with constant detail there…? It appears so, because the detail where the alpha was applied looks too big for 98K polygons. Using constant detail defeats the purpose of having a tool that allows for localized detail.
Try again but tesselating ONLY where the alpha is going to be applied, and I’m sure you’ll be able to have a lot more detail under 98k

EDIT:
It depends greatly on how you use dyntopo; here I used brush detail instead of constant detail and look at the difference:

Wireframe comparison:

[zeauro]

By default, Refine Method is Subdivide Collapse.
Your stroke subdivides mesh edges longer than precised detail but it also collapses edges smaller.
That is what is required when you define global shape of sculpt.

But if you want to detail by using a texture brush, it will be less problematic to jump to Subdivide Edges method and set Topology Rake to zero. Then, switch to Simplify brush and redo a pass to decrease polycount.

Brush Detail method was created to have detail consistent with a Brush using a texture with a View mapping. That way, you can set a detail percentage relative to texture details and you will obtain same desired result whatever brush radius is.

Whatever Refine method is used ; increasing a little bit Sample Bias when using texture can help, too.

[CLuss]

“The association between digital sculpting and performance is just marketing”
not so sure about that…