Sculpt, Retopo, with Displacement?

[Bao2]

A curiosity: MultiRes perhaps?

[brkn]

it's definitely possible, you'll probably need to bake a normal map. See: http://www.blender.org/development/release-logs/blender-246/render-baking/

[CGIPadawan]

Yes, in theory, I do know the solution lies in Normal Mapping and Displacement Mapping.

But right now in my experimentation, it seems Normals and Displacement are directly affected by the geometry and subsurf levels on the Lo-Res Model.

I have no problem baking and subsequently producing the Normal and displacement Maps. And I know they are correct because I can see the depressions (blacks) and the heights (whites).

The problem is the use of Re-Topo with a Lo-Res Model.

I read somewhere that Blender's ability to effect Displacement on a model is directly connected to the APPLIED subsurf level on a Lo-Res Mesh..

This would of course defeat the purpose of even having a Lo-Res Mesh (Weight painting even Subsurf 1 is a nightmare... you can do one actor this way, but try doing 20). [The other curiosity is that when doing SS3->SS1 baking it doesn't seem to be necessary to apply SS Modifiers on the SS1 model at all].

Now the thing about Re-Topo is that you can end up with topology that is quite different to the "sculpture" because when Re-Topo-ing one is thinking about animation.

However, the theoretical expectation is that if one has the Normal and Displacement maps baked over the Re-Topoed UV.. you can use Disp Modifier and Normal Mapping and instantly reproduce the illusion of the sculpted Hi-Res Model. Or at least that was my expectation.

This is where my experiment is falling apart.

Again, so far, I've only succeeded when baking Subsurf3 model details to a Subsurf1 duplicate. Now, I think if I can't get around this, maybe it would also work by heavily sculpting the SubSurf3 model (or even passing it through a Mudbox session) before trying to bake to a SubSurf1 duplicate... but that kind of workflow is a bit.... problematic.. and it really makes a hassle of trying to correct things once you have passed the Baking stages.

Again, this has not much to do with the baking (although maybe particlar Dist/Bias/Margin settings probably also come into play). But it's more about how important is it that the Lo-Res Mesh has geometry comparable to the High-Res mesh.

I mean, going back to my RE4-Leon example.. that thing is chockful of tris which doesn't sound like a good idea... and yet it looks okay on-screen. My concept is that you can end up with that kind of geometry if you try Re-Topoing to produce a Lo-Res model. My predicament is that after doing that... the Displacement and Normal Maps do not seem to work as expected (they cause rounded deformations that only vaguely follow what the pattern on the sculpted model was).

To SIMPLIFY the question it is just:

"Is it possible to produce a 'correct' Lo-Res Model by way of freehand Re-topology of the Hi-Res Model or is the only way (or best way) to produce Lo-Res model just making sure you had a SS1 version of the model before sculpting?"

Any ideas?

Cheers!

[LetterRip]

Displacement map actually displaces geometry and depends on the SD level, normal map is a lighting effect and should only depend on the resolution of the map. However extreme differences between the base geometry and baked geometry will give poor results.

Also the center location you are using with your retopo might be wrong (and is easy to get wrong unfortunately). So using your low res and baking from it will give you better results because it is certain to have the same center point.

Check this tutorial to be sure you are getting all of the steps done correctly,

http://www.katsbits.com/htm/tutorials/blender-baking-normal-maps-from-models.htm

LetterRip

[CGIPadawan]

thanks that was very informative.

Essentially, based on the tutorial, the short answer is that it is usually best to derive the low-poly versions from a "cage" or a duplicate from which the Hi-Res Model was derived.

It does make me think about using a Cage at SS3... to create an SS2 low-res model (with modifier unapplied) and to create an SS4 Hi-Res version that then undergoes a detailed sculpting process....

Making sure the "Cage" or "Mother" mesh remains allows to correct problems spotted later.

Also, as an expert-pro tip, the tutorial totally nixes the idea of using Re-Topo to create a low-poly mesh. So it seems that for now, Re-Topo is best abandoned or limited for use with the "Cage" or original mesh before it is replicated.

Thanks.

[fktt]

instead of re-topo I'd rather use the shrink-wrap modifier where applicable. :)

[CGIPadawan]

Well, my theory right now is to model one mesh first at SS3 to create a basic shape. And then once I'm happy with the shape.. make one copy at SS4 where i really pour on all the detail I can handle.

Another copy at SS2 will be for UV and should serve as the recipient of Displacement and Normal Maps.

[CGIPadawan]

Ok........ I'm kinda stumped with this. (Note: Blender 2.48)

So I made a basis mesh, then did a Hi-Res duplicate at SS 4 adding details, and a Lo-Res duplicate at SS2.

I made sure to add some SMALL details in the SS4 so that I can see if something worked.

So I bake an AO map, Normal Map (with Tangent), and Displacement Maps. All Selected-To-Active style.

The AO Maps work great as textures (set Color,Spec with Multiply instead of Mix)

The Normal Maps.. do nothing....setting MapOut to NOR doesn't seem to do anything.
(which is odd. I'm sure it worked before).

The real problem though is the Displacement Map.

Disp Maps don't seem to work unless the Displacement Modifier is below the SS modifier and I can't see a result unless I add SS levels to the Mesh (Which is exactly what I'm trying to avoid by making a Lo-Res Model!) I actually had to add in SS5 (even higher than the Hi-Res model!) To really see the Displacement.

And what tends to appear is very craggy looking Displacement.

I've included the file so others can play with it and tell me if I am missing something:

http://rapidshare.com/files/270365231/16ToothGear_HelpHelp.blend

Notes:
Basis on Layer 1,
Hi-Res on Layer 2 (with SS4 already applied),
Lo-Res on Layer 3

Both the Hi and Lo res started as duplicates of the Basis on Layer 1.

Thanks.

[Bao2]

Downloaded. I will take a look deeper later after Formula1

Your object in layer 3 has normal map and displacement. I suppose this is for testing both things and your idea is not applying both to an object because that would not have sense:
normal maps is for games and displacement is for production. The two create an ilusion of high poly but the first one don't create displacement so you see the edges are not displaced like in a bumpmap you can see in the edges of an object it is not displaced. Normal map is for games because it uses low poly meshes.
Displacement is used in rendering time and adding a displacement modifier to the object is not needed (perhaps you did to visualize).

Another thing: In Blender choose File / Compress File and then when you save you have a 8 Mb file instead a 25 Mb file.

As I said in post 2, what about using MultiRes instead Subsurf? When you sculpt in a high res your vertices in the low level move too, and I think it is better for displacement maps than doing the high poly sculpting in a subsurfed model (where when you sculpt you move vertices but your lower mesh ones not so you could have a scenario where you have a hill of vertices in your lower mesh but is is a valley in your higher mesh and it probably don't works optimally that way). So you have a multires and when you finish your sculpting make a copy and apply the modifiers and you have the high poly and in another copy you go the low level and apply the modifier and you have the low poly. But in both the vertices are in similar positions so the displacement map is optimal when calculated in my opinion.

I will take a look later.

[Bao2]

Sixth place in Formula1 for our Alonso (in classification he had a broken brake) so it is interesting see next week in Belgium.

Well, forget Formula1 and back to our dear blender.

I was looking and the main problem is that the unwrap is not well done. In a correct unwrapping you can see the faces extended by the image. In your "unwrapping" the vertical faces of the object don't show in the unwrapping. So first thing to do is create a correct unwrapping.

I will continue and will post a step by step or a file.

[CGIPadawan]

Hello Bao,


Thanks for the quick reply. Yes I am aware the UV unwrap is not good. But I was more keen on using this as testing only.

I think even with this "bad" Unwrap, I would have at least seen the effect of the Normal and Displacement maps on the circular pegholes and the circular rivet/bump on the five sided mounts that extend at the center of the gear cog.

My plan was to see about using Normal maps for small details, but Displacement for big details and I wanted to compare their effects on this mesh. (Again, even if I know the unwrap isn't good.. I was just looking for those circular effects since they represent detail present in the Hi-Res model that isn't in the low-res model).

I will try Multi-Res modeling for sure... Thanks for that advice. :)

[Bao2]

Well, here we go.

A youtube HighQuality video: http://www.youtube.com/watch?v=WXJMnYTE4Bk

And here is the text you see in the video:
with step by step instructions for Normal map:

Open the file 16ToothGear_HelpHelp.blend
Arrange/Join the windows so you have only these windows: a 3Dview to the left, an UV/Image Editor in the center and an outliner to the right. The buttons window below the three.

In layer 1 change the name of the object to Gear1 instead 16ToothGear_ORIG
In layer 2 change the name of the object to Gear2 instead 16Tooth...
In layer 3 delete that object (we will use only the other two). Move the lamp to layer 1

In object Gear1 delete the modifier subsurf.
In object Gear2 delete the material.
In the outliner select camera and in Editing panels (F9) in the Camera Panel change Lens from 35 to 67. Select layer1 and F12 to render. That is the low poly version.

Select Gear1 object and enter edit mode and select all the vertices. Press U and choose "Unwrap (smart projections)". Later you can try other maps like the one that was on the object in layer3 but I didn't liked it because all the vertical faces of the object are missing and we could want have later sculpted detail there and it would not show with such a uvmapping like that object in layer3 had.

Well now you see in the image editor the uvmapping. Don't create any image yet, we will do so to learn how to avoid a silly error.

Exit edit mode of Gear1.

Now in the UV/Image window go to Image menu and choose New and give it the name: NormalMap. Now Image / Save and change the name from Untitled to NormalMap.tga and save the image in the same folder as your blender file.

Go to the scene buttons (F10). There you have the Bake panel. But before we need go to the outliner and select Gear2 and hold Shift and select Gear1 so we have the two selected and Gear1 is the "active" object (the last one selected).

Go back to the bake panel. Make selected if they are not, the "Normals" and the "Selected to Active" buttons. Choose the "Tangent" method. Click BAKE.

It gives an error: "No images found..." and this is the silly error I wanted to show.
In the UV/Image editor you can see NormalMap.tga selected but we didn't assigned it to the faces. To do it enter edit mode in Gear1 object and select all the vertices. And now in the UV/IMage editor select NormalMap.tga in the list to show again.

Exit edit mode of Gear1.
In outliner select Gear2 and hold shift and select Gear1.
Press BAKE and a blue image appears but it clearly is not well done. And that is another silly error.

What the hell is happening now???
Well, make layer1 and layer2 selected (click on layer1 and hold shift and click on layer2). Now the two objects are visible in the viewport and that is essential for the calculation to occur.

Look in outliner to have selected first Gear2 and last Gear1 as said and
now press BAKE and the correct normalmap appears and is saved to disk.

Unselect layer2 so only layer1 is selected. Click F12 and click J on the rendered image to it to disappear and a new buffer be available. Don't close the rendered window.

In the outliner select only Gear1.
Add a material to Gear1. To the right you have the Texture panel, click Add new to add a texture. Two new panels appear: MapInput and MapTo

In MapInput change Orco to UV instead and just above there is a edit box "UV:" and enter in it UVTex

In MapTo unselect Col and select Nor. In the Nor:0.5 below change it to 1.0

Go to Texture buttons (F6) and in Texture Type select Image. In the MapImage panel select "Normal Ma". In the Image panel open the list and select "NormalMap.tga"

F12 to render and click J in the rendered image to switch to previous render and compare.


You are in the contrary side of the world that me so I will be in bed while you try all this. Tomorrow I will read if any other doubt about is yet. :)

[Bao2]

I forgot to say that the unwrap I did with "Unwrap (smart projections)" is also bad because we have two small circles and enormous rectangles of the vertical faces. I did it to demonstrate that it is needed to create a uvmapping that have all the faces expanded in the map (avoiding overlapping). The correct would be make the two circles much greater and the rest of faces much smaller to their real sizes. Just some B to select and S to scale and in some seconds you have a much better distribution of the uvs.

I didn't tried but I think you are right that with this object your uvmapping would work also because where the normal mapping is needed you mapped. But I don't know what will happen with the vertical faces, perhaps some artifact with the calculation would occur? I didn't tried.

[CGIPadawan]

Hello Bao,


I was just reading up on MultiRes modeling with Blender and it definitely looks appealing. But you were saying something about how SS does not affect lower-level vertices.

But... won't it be all the same once either SS and Multires is "applied" permanently to a model? (ie: The SS or Multires is turned into actual geometry).

Or are you saying that a Lo-Res Model with Multi-Res will achieve better results than one using SS?

I plan to start a new exercise on this using something that might be easier to UV and appreciate... The Mil-Mi24 Hind Helicopter model (but without the transforming ability).
I don't want to lose time UV-ing the Gear as I want to focus on studying the translation of details from Hi-Res to Lo-Res Models. (Mostly to save time.. Though I've read some worrying sources saying Blender's Normal/Disp mapped models will not retain detail in closeups or under certain lighting conditions).

For sure though I'll start experimenting wtih Multires. If it makes "virtual" subdivision and I can still work in Lo-Res Weight Painting/Rigging if the Multires remains unapplied, I may opt for a strategy where both the Hi-Res and Lo-Res models have Multires levels. If Multires Propogation from Higher to Lower Levels is very good, then perhaps my Hi-Res Model must be used to derive a Lo-Res model that is just the same model with a lower Multires level (deleting higher levels and relying on maps derived from the Applied Hi-Multires model). This is PLAN-B.

-----

Sidenote: Alonso did pretty well in Valencia didn't he? But wait till you see what he can do when he goes to Ferrari. That team is a team in need of a demanding driver of Alonso's level. Kimi had his chance, but it seems he is not as dedicated as Schumi was or Alonso is when it comes to demanding performance from his team/designers/engineers, etc.

[Bao2]

I realized I copy and pasted bad the latest part so I corrected the above post.
It is insanely late here so I go to bed. Tomorrow I answer your last post.

[CGIPadawan]

Hello Jose,


Just saw your 2008 Renault project thread. That work is very good!
That's exactly the level of detail I want to "fake" on something with fewer polygons....

Let's take your car for example... what MultiRes level was that? If it's level 4 for example.. can it project on MultiRes Level 2 with Normal/Displacement?

Thanks! And... wow.. nice car!

[Bao2]

I wanted to create in Ubuntu 64 a capture session and then post it in youtube and
what better than try with the instructions of post 13 of this thread. I did not add sound but a text file where I show what I am going to do. It was just a test to see if youtube will show good quality and it don't loose too much detail when youtube recompresses it. See it in HD and if you uses like me the free DownloadHelper plugin for firefox you can download (HQ 22 is the high quality one). This is the result:
http://www.youtube.com/watch?v=WXJMnYTE4Bk



I was just reading up on MultiRes modeling with Blender and it definitely looks appealing. But you were saying something about how SS does not affect lower-level vertices.


Yes I was wrong in that. In SubSurf the lower mesh is also affected I can see now.


Though I've read some worrying sources saying Blender's Normal/Disp mapped models will not retain detail in closeups or under certain lighting conditions).


Normal Map is a light effect in rendertime. It don't displace vertices but only changes the intensity of the bright of the pixel based in the normal map. So with no actual displacement of the mesh you have certain limitations. It is good for a game but it doesn't look good for a movie or so where you want actual displacement of the mesh.
I was today trying make the displacement map and i have wrong results. Probably displacement is not suitable for such a object like this gear so I will test in a more organic mesh to see how it works. If you have any problem post here and I will do also if I am unable to do displacement. I am learning Blender and I don't know exactly what the limits are and how to do all the things I did in max.


For sure though I'll start experimenting wtih Multires. If it makes "virtual" subdivision and I can still work in Lo-Res Weight Painting/Rigging if the Multires remains unapplied

MultiRes seems very good but I also have to test more and find the situation where it is better than Subsurf if there is any.


, I may opt for a strategy where both the Hi-Res and Lo-Res models have Multires levels.


I think MultiRes is to avoid using displacement maps and normal maps. You select the level you want to render and so you low in the objects that are slow in the render and up with the objects that are close to the camera and showing the high details. So it seems a tool better to rendering than to modeling as I was thinking too.


Sidenote: Alonso did pretty well in Valencia didn't he? But wait till you see what he can do when he goes to Ferrari.


He already signed for Ferrari. He will be in Ferrari next year (but he can't say it until end of this season). Go to Ferrari is his dream. I like him in Renault too because it is like a family.



Just saw your 2008 Renault project thread. That work is very good!
That's exactly the level of detail I want to "fake" on something with fewer polygons....


The model you see is around 60 thousand vertices. It has no "subsurf/multires" objects and seems perhaps more high that it really is. When I model I don't waste vertices in places where no one is going to see them.

I really would like to end it but I just now like more playing with blender and learning it.

[CGIPadawan]

Well it is possible to avoid using Normal and Displacement Maps when using Multi-Res Levels.
However, Multi-Res cannot be used with Shape Keys. And we all need Shape Keys for human faces... which is one of the highest applications for Multi-Res.

hehehe.

HOWEVER, doing a bake from a High Multi-Res model to its Lower Multi-Res duplicate should, in principle be the same as doing my usual SS3 to SS1 baking, except that Multi-Res models can be sculpted to higher detail with cleaner topology!

So maybe a solution is within reach!

Going to try playing with Multi-Res now!

[CGIPadawan]

Hello Jose,


I think MultiRes is great for producing "ultra" levels of detail from a Mesh modeled with high-SubSurf levels applied. Something like adding grills, rivets, and things to something with topology that is "evenly added" (because Blender is the one calculating additional Geometry). HighSS/LowSS combo still seems the best combination though because Multires details really lose a lot going one or two levels down.

Hmmm.... Would this produce better Normal/Displacement Maps?
I must try and also watch your tutorial :)

I will also test Multires in a rigging situation.... At least this might mean that using Multires for armor and other static things like that means those things won't need normal displacement maps and still achieve fast times.


Perhaps the REAL problem all along has been that Blender trips up when I "manually" add details to a mesh? And it cannot map them properly as normals/displacement?

Like you said, it wasn't working with the "gear" models. But I didn't think it should have been any different. :hmm:



He already signed for Ferrari. He will be in Ferrari next year (but he can't say it until end of this season). Go to Ferrari is his dream. I like him in Renault too because it is like a family.



Yes. I kind of figured it out when I heard Banco Santander was going to sponsor Ferrari. Santander is kind of like a banner for Alonso. To be honest, I was always a fan of Schumacher, but my real love is "hard work". And I have no doubt Alonso is a hard worker. I have been heartbroken too long with Massa and Kimi as they don't seem to have it in them. But I am sure Fernando will not let us down. :)

[CGIPadawan]

Okay... Your video was very informative. Seeing it I saw that I just missed a step.

The Normal Map problem was caused by not Pressing the "Normal Map" button in the Texture Image panels. HAHAHA! Whose idea is it anyway to require that you enable "NOR" in the MapTo AND press "Normal Map" in the Image Textures panel!

I got my "Ugly UV" version to also now show the normal map working. Thanks for the video.. and yes the video looks great. My normal map though looks ugly, but at least now I can SEE that it's working... even under bad UV conditions.

As for the Displacement... Yes I think I spotted it in your video. Definitely not what either of us had in mind! I got the same result... turning my 16 tooth gear into a "Mushroom/Flower" thing. :P

I'll try it on something with a more basic shape like maybe a Shield and Emblem as a hybrid of my learnings from you and my past workflow.

But for now.. Sleep.

Thanks. :)

[Bao2]

However, Multi-Res cannot be used with Shape Keys. And we all need Shape Keys for human faces... which is one of the highest applications for Multi-Res.


:shrug:


I will also test Multires in a rigging situation.... At least this might mean that using Multires for armor and other static things like that means those things won't need normal displacement maps and still achieve fast times.


MultiRes has a slot where you indicate the level in what you want the modifiers to act (bones in this case) so it seems great for riggings.


To be honest, I was always a fan of Schumacher


All of us are too!

[CGIPadawan]

Yeah.. I totally relate to your shrug about the ShapeKeys+Multi-Res/Sculpting incompatibility.... This is definitely a limitation for me.

Hmmm... interesting note about the rigging slots.... But maybe this week I will try something where I try to use Multi-Res sculpt to add detail that is intended to appear as Normal and Displacement Maps (instead of trying to add details in SS Modelling for Normal and Disp).

But I am really sleepy now.

Thanks for the help Jose... Hope this is also augmenting your learning of Blender.
It is a good goalpost to "match 3DS output" with a different tool like Blender.

Imagine the savings once you succeed! :)

Good night from the Philippines!

[FishB8]

Yeah.. I totally relate to your shrug about the ShapeKeys+Multi-Res/Sculpting incompatibility.... This is definitely a limitation for me.

I think this limitation is due to there being no way to bind the shape keys to a specific multires layer.

Both of these have to operate on the base mesh itself, so there is really no way (in the current implementation) for both to work at the same time.

I may be wrong, but I think in 2.5 these are both treated as modifiers, and there has been work done to allow modifiers to be stacked in any order so that no modifier has to have direct access to the mesh data.

[CGIPadawan]

Hello Reuben,


That is definitely interesting news.
But I believe that a "simpler" solution developed prior to the onset of the "new flexibility" brought in by 2.5 will still benefit me (and others watching this exercise) since it could allow for production of both Detail and Performance using low SS levels and maps.

For now though, it is somewhat strange as I believe ShapeKeys work with unapplied SubSurf modifiers. And yet as Jose is discovering... Multi-Res and Subsurf have a number of redundancies.

Going back on the topic, this morning, I've started a new experiment where I will use all 4 Hi-Res/Lo-Res applications: Actual Geometry, Sculpted Detail, Normal Maps, AO Maps with a project that has both interesting detail and easy UV layout:

A Set of Warhammer 40k Combat dice! :applause:
http://thehobbybox.biz/images/Gale%20Force%20Nine/91002CombatDice.jpg
Basis Image above.... :wise: not actual models... ;)

I figure the Skull is 6 and this is where I will use Sculpt Mode combined with Geometry for an engraved look. Maybe even use Multires/Sculpt to add things like microcracks...

The dots will be concave crevices made with geometry actual Geometry and I'm thinking about Painting a "6" on the skull for good measure.

I will incorporate my original Workflow with Jose's version of it, this time being VERY selective about where Sculpt can be done and what details will be left to Normal and Displacement Maps (since the previous experiments have shown it is not that effective in bridging Geometric differences)

The new workflow will be:
1) Model Dice in SS3 to as high detail as possible without making awkward edgeloops and bad topology. (Probably to include the engraved silhouette of the Skull).
2) Make copy of the Dice in SS1 (this is Lo-Res)...but using Render Level 3 (my original workflow)
3) Apply SS3 to the Hi-Res Dice
4) Put Hi-Res Dice into Multi-Res/Sculpt (to add details to the Skull's eyes, nose, teeth)
5) Start Baking
6) Cross fingers. :curious:

If it works, I'll use some Materials from the Blender Material Repository which should do nicely for this project. And we can have comparative renders.

Stay tuned until the end of the week to see if this works. :p I know SS1 seems a tad low... but I really want "fastest performance" from my actors when the time comes.... I can worry about closeups later. :twisted:

Ciao for now! :)

[Bao2]

I was about to ask if you was using luxrender in that render.
Glad to know it is a photo.

As an exercise well, but of course in real life the thing would be modeled this way:

Create a circle mesh. Create five copies and rotate to their positions. Create faces to join all the circles and then select all the faces and make two or three cuts more and move to give a rounded look.

Create a plane. Make a hole with all the detail you want. Now we have a hole in a plane. We need to move the hole to the mesh and cut a rectangle in the mesh and put our rectange with the hole instead. Make this many times for all the holes.

For the skull create a white and black image of the skull. Make it gray instead white and black and blur a little to avoid the displacement artifacts in the wiki page are explained when using displacement maps. Then apply as a displace modifier or a displacement map what gives the best result.

Then unwrap and paint color and dirt maps. Go to luxrender and post here !

[CGIPadawan]

Hello Jose,

Thanks for the feedback.

I am doing it slightly differently, but using a subsurfed cube subdivided twice and with SS Level-3.

And then I will subdivide the top face (skull/6) to try and produce the skull as a combination of geometry "engraving" for its silhouette and then use Multi-Res Sculpt mode for the details for eyes, nose, teeth.

I think, though, your method is faster, but I am halfway through it already.

But thanks... and yes I will post here the rendering results.... or the problems.. if I get problems.

[CGIPadawan]

Ok, so now here is the mesh so far,

much of this the Lo-Res Model can inherit:
http://img259.imageshack.us/img259/4213/dicemesh.jpg

Tomorrow begin working on the skull engraving, but first I must find a nice picture of a skull.. and then use Limit Viewport (ALT-B) so that I can neatly model the Skull Engraving over it just like I'm following a set of blueprints!

Sleep again. :)

[Bao2]

From your image in Gimp with cut tool, perspective tool, resize image, convert to greyscale, contrast and blur:

http://yfrog.com/4246391155j

Another version with less filters:
http://yfrog.com/4269849590j

I blurred them a little because in blender wiki manual it says it is needed to avoid artifacts.

[CGIPadawan]

wow.. that's pretty amazing... but I was going to try this image:

http://www.nlm.nih.gov/medlineplus/ency/images/ency/fullsize/8915.jpg

You know... so I can use Sculpt to maximum effect... hehehe

Won't follow this exactly (the Skull has to be flat afterall with just a "canal" going around it for the "engraving"). But I am interested in using sculpt mode for those "wedges above teeth"... and some of the chin and temples detail.

We shall see.

But remarkable work extracting the skulls from the actual dice... :D

[Bao2]

I can't wait to see that dice sculpted. I am very curious about it.

[CGIPadawan]

Yeah? Well we'll all have to stay curious a little longer.

I'm having these nasty "bumps" around the skull because of all the triangles that appeared when I subdivided only the section around which the skull will be engraved.

In subsurf these tris cause bumps. I think Sculpt can fix this, but I'm not risking it.
(I did try using Smooth [Select Vertices, W -> Smooth] and it causes "erosion").

I will try to save the topology by simply deleting the tris and try to flatten the thing out with quads.

If that fails, then I will revert to Jose's solution: Circles for the Dice Faces. :p

To do this I will have to make the circle-faces as a separate mesh first, model with X-Axis-Mirror, and make a sort of "chopped sphere shape"... then use Shift-Ctrl-S to "Make Sphere" so that I have a nice "sphere that's been cut with 6 circular hole faces". Then join the dot-holes and skull silhouette to this mesh and "connect the dots"....

...THEN apply SS 3. :P

btw: One of the neat things about SS over Multi-Res is that you can select vertices and press Shift-E and control how much Effect Subsurf has on a section. So you can do "rounded cubes", and other shapes more easily by using SS+Shift-E than for example sculpting such in sculpt mode. Just like in all other SS situations, the underlying vertices follow the Shift-E influence.

After this topology problem is overcome I will go into the sculpting....

Geez.... I really have much to learn about modeling. :shrug:

[Bao2]

Geez.... I really have much to learn about modeling. :shrug:

Modeling is faster if always try the quick methods first. But in your mind not in the computer. So I always see if the thing can be done with simple shapes, or lofting a shape or extruding a shape or so.

In this case beginning with the circles is almost done. If you start with a cube or a sphere it is not so easy to model.

Shift + E for creating the creases is something I use too. But in max we could see the number and I miss that in some other areas too (like in painting weights of vertices when skinning, in max we can see the value of each vertex weight to each bone and I suppose in maya and others apps too). I don't understand why in Blender they don't expose the values of things like these.

[CGIPadawan]

That is a good point. :)
Thanks.

I think your point is better put as "Think of the 2D shape you can extrude or cut quickly to produce something".

Actually I am more accustomed to "extrude" modelling where i start at say the nose of an airplane, or the lips of a head.. and keep extruding and "growing" the mesh until it covers the whole.

Modeling from a "basic shape" (like in this one where I tried a cube) is always problematic for me. That's right, I can do a helicopter and a human head, but a dice gives me trouble.. hehehe.

Again, I am holding on to that "Use Circles as Faces" solution as PLAN-B if I cannot save the topology around the Skull.

Thanks again. :)

[sebastian-koenig]

hi!
just an idea: why don't you just use a displacement map in the first place? modeling these details would be quite tedious, i guess. so, my idea was that you create one version of your mesh with quite a bit of subdivisions (not modifier), so that you have a rather dense and even distribution of faces.
then add a subsurf-modifier to it, set it really high.
add a displacement-modifier and create the texture for it.
use a black and white image with a high resolution for the texture.
then you probably have to adust the strength of disp-modifier.
but it should give you the amount detail that you need.
then bake this to the low-res-version of the model (of course assign an for that image before).
http://img.skitch.com/20090826-n9mcjw28bgm9tfgs5t8gfrpsny.png (http://skitch.com/sebastiankoenig/bhaut/blender-users-sebastian-01-blender-blendfiles-int-tests-modeling-displacement-disp.blend)

http://www.blender.org/development/release-logs/blender-246/render-baking/
http://wiki.blender.org/index.php/Doc:Manual/Textures/Maps/Bump_and_Normal_Maps
http://wiki.blender.org/index.php/Doc:Manual/Modifiers/Mesh/Displace

of course this may not be what you want, but it could save a bit of work ;-)

cheers!
seb

[CGIPadawan]

Very good Sebastien.

And you are correct about using Displacement on Hi-Res Meshes.

The specific challenge Jose and I are grappling with is "Displacing a Lo-Res Mesh".

Or rather... not painting in the detail.. but seeing if details done in Hi-Res can be transposed to Lo-Res Meshes by "trickery" (normal maps/Displacement Maps).

In the first post, I mentioned RE4 (which I am still playing in order to peer at it). And somehow they've managed to do it. Okay.... It's not Pixar-quality... (And Leon is a TERRIBLE actor in that game!) But I wasn't saying it had to be Pixar-ific. I wanted to see if Actors can be made in that image though.... Lo-Res posing as Hi-Res complete with ShapeKeys (which disqualified Multi-Res as an option) and rigging and deformations.

Although we know doing what you are suggesting (Painting a Displacement Map for a Hi-Res Mesh) will obviously work well with our Dice face... We are specifically trying to make a Hi-Res Dice project on a Lo-Res Dice.. hence all the trouble we are going through for the Geometry, etc.

You are welcome to join in! :applause:

[CGIPadawan]

Oh drat! That's what I get for not reading thoroughly.

Seb! That is a good strategy!

I was really thinking about something like this in the beginning!

But the whole time I was trying to force a Lo-Res Mesh to accept displacement!
And force a Hi-Res Mesh to accept actual geometric detail!

So you are saying I can DISPLACE the Hi-Res Mesh... and BAKE it to a Lo-Res Target!
That is great.... for grills.. Lines in the hair.... Gills on a shark....AND possibly this Skull thing.

But I suppose if we are talking about things like "Pegs", "protrusions"... then that's where Geometry should be applied...

Interesting.

Yes.. I think I try your approach first.

P.S.: But this means both the Hi and Lo Res Models must have UV data right? To use the Displacement Map on the Hi-Res and the Normal Map on the Lo-Res.

P.P.S.: The other curiosity is.. If I press APPLY on the Displacement Modifier... does that mean the changes become actual Geometry? And then I can use Sculpt to add details?

[sebastian-koenig]

hehe, yeah, that's what i have been suggesting :-)
yes, in this case it means that both have to have UVs. if you just wanted to bake the normal or displacement from a sculpted highres mesh, you only need to have the lowres to be unwrapped, with an image assigned (btw, i suggest 4096x4096).

as for the characters: normalmap is really the way to go. i think every game uses them to get the details on the lowres-meshes, and the workflow should always be the same: create character, duplicate, move to different layer, add multires, sculpt and then bake this back on the basemesh. that is the same for normal, diplacement and occlusion-map. (although you have to use a 32bit image for displacement)
shapekeys might be a problem when you are using a displacement-map, but not necessarily. i mean, if you have shape-keys, you just transform the mesh, and thus the UVs too. so the displacement-map will follow, and that should be the same for normal-maps.
i would suggest to use a combination of both: displacement-modifier for the not so tiny details and the overall shape-definition, normal-map for all the smaller details. then you only need a subsurf-level of 2.

P.S. yes, apply will apply the changes to the geometry. but then you first should apply subsurf. which, by the way, has of course to come first in the modifier-stack :-)
you can sculpt at any time and any level, but it makes no sense to use subsurf-mod here. only if you apply that too, before sculpting. but i'd always suggest multires for sculpting details.

P.P.S. if you want to go REALLY fancy, i think it should be possible to blend images for the displacement-texture using nodes. didn't try that though. but the idea is, that you assign a node-based texture as disp-mod and for every shape that you have you bake 1 displacement-map. in the nodes you can activate the image using time-nodes, so that you have a different displacement-map for each shape. maybe you can even activate them automatically using python-nodes, but to be honest, this is totally over my head, so i better stop talking now... :D

[CGIPadawan]

Yes yes... you are saying exactly what I had hoped to be doing... but not really for games... I was thinking about doing this for a movie next year...

Basically the idea was to do "scale" scenes.. like large geodesic domes... foliage.. AND an actor hunting a creature...All animated in Blender.

Now I know that using Compositor you can get away with all of it....even if each basic part of the scene approaches a high number of verts. It just means you render all of them separately using the same camera... and then you force-blur parts in Compositor to fake a DOF.

But the more you can do "live" the better right?

I guess the lines of defense would have to be:

Try Actual Geometry... if it gets messy....then....move to Displacement Mapping + Baking.. and if you want the really tiny details... Normal mapping + Baking.

I suppose with time... one will gain the knowledge of whether something should be sculpted.. or done with a Displacement. ;)

[CGIPadawan]

Ah.. definitely you have touched upon something interesting. :)

At least I know what I have envisioned is in fact possible. Only that my meager knowledge kept me going at forcing the use of Displacement maps with Lo-Res models. I mean.. if it really worked that way then it would be amazing. :P

I should have known it's the Normal maps (combined with AO maps and correct lighting) that have always been intended for this purpose for the Lo-Res Maps and that the Displacement Maps achieve hard-to-mold details for the Hi-Res models.

I suppose Normal Maps with Ambient Occlusion mapsand Specular Maps can combine to create the impression of very high detail. Perhaps the use of 4,096 size maps can even do a few a things that will stand up to close inspection?

And yes. A Lo-Res model at SS 2 can probably carry all that.

Finally, light at the end of the tunnel. Thanks. :D

[sebastian-koenig]

sounds interesting!
but here's what i think:
use displacement and normal-map all the time, right from the start.
so your workflow would look like this:
1.model your basemeshes. go for clean and efficient topology.
2.create highres-versions of them using multires and sculpt them.
3.bake the disp- and normal-maps for each of them
<-- advantage over using the actual high-res-geometry: your scene-and animation-files will be much smaller! viewport more responsive
4. when you animate you can decide whether or not to use displacement-mapping or not and at which level. for example: objects near the camera use disp-mod+ subsurf level 3, objects far away use no displacement at all and low subsurf.

i'd say, using a normal-map doesn't really hurt, but for objects in the distance i think you don't need it. and of course, DOF can help a lot. So using the defocus-node and a dof-target-object (e.g. an Empty) is probably a good idea.

for things like the dice i would just bake a normal-map by using the technique i explained and be done with it. or, if you need a closeup, use the displaced highres-version of it.

[CGIPadawan]

One of the beauties of the method you are describing is that it is actually possible to build ultra-high details in installment.

For example.. with the cube you showed.. You can make one Displacement Map that contains that squiggly pattern..... Try it... APPLY....
Then make another Displacment Map that contains another detail... try it.. APPLY.

When it's done the High-Res Model contains a TON of detail that when baked to Normal Maps or Displacement Maps, AO.. should work well with the Lo-Res Model.

Because in theory you can just add and add.. and test renders.... It's like painting a house with repeated coatings. :)

[sebastian-koenig]

One of the beauties of the method you are describing is that it is actually possible to build ultra-high details in installment.

For example.. with the cube you showed.. You can make one Displacement Map that contains that squiggly pattern..... Try it... APPLY....
Then make another Displacment Map that contains another detail... try it.. APPLY.

When it's done the High-Res Model contains a TON of detail that when baked to Normal Maps or Displacement Maps, AO.. should work well with the Lo-Res Model.

Because in theory you can just add and add.. and test renders.... It's like painting a house with repeated coatings. :)
yeah, right, but why apply?
you can just use the modifier and bake with it. no need to apply.
hint: you know you can use more than just 1 displacement-modifier, do you? and that you can limit the influence of it to a certain vertex group... and that you can setup up to 8 different UV-textures, each with it's own name and layout. so you can use 8 different disp-mods with a different uv-layout each, and with their own vertex-groups...
there are indeed quite a few options here to use....
;-)

[CGIPadawan]

yeah, right, but why apply?
you can just use the modifier and bake with it. no need to apply.
hint: you know you can use more than just 1 displacement-modifier, do you? and that you can limit the influence of it to a certain vertex group... and that you can setup up to 8 different UV-textures, each with it's own name and layout. so you can use 8 different disp-mods with a different uv-layout each, and with their own vertex-groups...
there are indeed quite a few options here to use....
;-)

:p

Now I know how Steven Spielberg felt when he was in meetings for "Saving Private Ryan" and he was going through excitedly about how he discovered this amazing way to shake the camera for shooting the explosions by attaching the camera to a hand-drill.....

And then he's interrupted by an experienced Cameraman who told him: "There's this thing we use now called a Lens Shaker for filming explosions?" :wise:

... And that was the end of Spielberg's brilliant "Hand Drill Camera"

;)

[Bao2]

I think your point is better put as "Think of the 2D shape you can extrude or cut quickly to produce something".


Yes. Imagine you need to model a gun for example. The way I will do will be: Go to a side view and draw the contour. Then go to front view and extrude. And then just refine here and there and it is done.

What I am using now in Blender and I was not using in max to model is the sculpt tool. I love the sculpt tool to push/pull the vertices one to one in the low poly mesh instead grab them in areas where I see problems.

[CGIPadawan]

Yes... Definitely should learn that too.

Thanks. :)

I think I will try Multi-Res and apply.

And then I'll be using Seb's method for "one pass of Displacement Mapping" and APPLY using your version of the Skull (because it's a perfect two tone image). We can then experiment with what level of Displacement is good.

APPLY.

Then use Sculpt mode to try adding details.

And THEN we try Baking.
For the simple Dice test I expect this method to work.

[CGIPadawan]

Ok,


I went back to one of my backup saves before I took an axe to the model and caused all those tris.

Made two copies of the Base Dice Model (SS3 unapplied):
1) Hi-Res Model (SS3 Applied)
2) Lo-Res Model (SS2 Unapplied with Render Levels 3)

Both have the same UV coordinates.

I took Jose's Black and White Skull (with less filters) and made a Displacement map for the Hi-Res Model to get around the hard work of sculpting, mesh-wrestling.

And this is the result:
http://img405.imageshack.us/img405/2683/skullengraving.jpg

To make the Displacement look really nice, I ended up adding Multires = 4.

As for the Displacement Modifier... Midlevel is the level of color at which there is no Displacement ("flat") (Black = 0.0, Grey = 0.5, White = 1.0). Darker colors relative to this cause lows, and colors lighter than Midlevel cause peaks in Displacement.

So using Jose's Skull extraction, I used White as the Midlevel. Strength indicates how deep the engraving is and in this case it's just 0.1 because Jose used a solid Black for the lines.

Earlier it must be noted Sebastian mentioned using multiple channels/multiple textures and this is where something like that comes into play because one can paint multiple Displacement maps of various details and use varying degrees of Displacement. But for this exercise, I only use one.

Now... What is to happen next?

I will try pressing Apply on the Multires and the Displacement. And that commits all the detail to Actual Geometry. This means that a basic Hi-Res model may be possible at SS3 + Multires4... for Sculpting/Displacement work. At this stage you can use more channels, or start Applying and clearing them out if you run out of channels.

Once all that is Applied as actual Geometry on the Hi-Res Model and there's nothing left to add, it will be time to Bake all of it to the Lo-Res Model.

The earlier experiment with the 16-tooth gear shows we expect ready success simply by using AO maps on MapTo->Col (Multiply) with Normal Maps. My previous experience with trying to bake Displacment Maps to Lo-Res also yields a "grey map" that is very "mild" but could also help nudge detailing on Lo-Res Models higher. Might be nice to try for this excerise.

Then it will be time for some renders. :)

More Curiosities!

1) In Blender's page about Displacement Modifier it shows an animated Slime Demonstration using Displacement Modifier. Is this what happens if an Image Sequence is used as a Displacement Map?

[sebastian-koenig]

hey!
l think it is look looking good!
the animation you see on the blender-site are multiple displacement-modifiers with procedural-textures (i think just a regular clouds-tex) with IPOs that animate the offset. probabls the z-offset.
:-)

[CGIPadawan]

I see. So in that way it would also be possible to do "sliding floors" by manipulating one off-set on the textures.

And that also means Applying the Displacement Modifiers would destroy the animations because the geometric changes will no longer be joined to the Textures but grafted to the Geometry at that specific point in time in which Apply was pressed.

Oh and btw Seb, I sent you a PM. :wavey:

Thanks for the compliment.. The next phase is to try Apply and Bake... And once all the Geometry is done....(including baking-and-faking). We will move on to doing some interesting things with the colors.

I am thinking it should be possible to also use say.. Node-based multiple materials on the Hi-Res Model and Bake a map for the Lo-Res model that employs all the colors that were used from the Blender Material Repository....

So then we can have fun playing with Repository-sourced Marble, Copper, etc....

But even if that doesn't work.. combining GIMP-painted textures with the use of an AO Map and Spec Map should actually still work.

Like I said earlier.. it's working. :)

[Bao2]

Looks good.

What renderer do you are going to use? What about trying luxrender?

[CGIPadawan]

Looks good.

What renderer do you are going to use? What about trying luxrender?

Actually Blender's Internal Renderer is quite good. There are a few good results that will surprise you. Having said that, of course other Renderers can be better in some areas...

But I will try using Blender Internal and really play around with it then maybe try with Luxrender or other renderers. But I'm not so fond of experimenting with those.

[CGIPadawan]

Ok.. I've successfully baked the following:

1) AO Map
2) Normal Map
3) Displacement Map (with 32 bit Float).

Blender kept crashing though.. and just navigating around the Hi-Res model had become VERY difficult.

Blender's performance does improve if one reduces the number of Viewports.

But it crashed every one or two bakes.

Got all the maps though, so we can see first if the Lo-Res Model reacts well to this first set of maps. Then we try Sculpt.
:D

[Bao2]

But it crashed every one or two bakes.


I was just thinking today that in this month that I am dedicating to Blender, never crashed.
Perhaps you have a bunch of vertices in the same position? I would try remove doubles and of course flip normals out too.

The main reason I am trying blender is because luxrender. I love these "real photo renders" because they are the "make art" button. You create a light and a cube and a ground and it seems real with an unbiased renderer.

[CGIPadawan]

Ok....

Here's what happened after baking and all these maps are used:

Original Hi-Res:
http://img34.imageshack.us/img34/176/hiresrender.jpg

Lo-Res with AO and Normal only:
http://img141.imageshack.us/img141/83/loreswithaonormonly.jpg

Lo-Res With AO, Normal, and Displacement maps:
http://img440.imageshack.us/img440/9870/loreswithdisplacementao.jpg

Observations:

1) I really have to stop using Displacement Maps on Lo-Res Objects. :hmm:
2) The Hi-Res mesh really started causing a lot of Blender crashes:
a) When trying to select even one vertex it causes a crash
b) When trying to use Decimate it says "Non-Manifold Mesh"

Resolutions for next experiment:

I think I broke something when I applied MultiRes 4 over an applied SS3 mesh!
Perhaps I should try producing another Hi-Res Mesh from my Base Mesh, and this time removing the SS3 and just trying MultiRes 4 or higher by itself. I am still wary if whether Apply Displacement caused the Non-Manifold problem.

If I abandon using Displacement maps for Lo-Res Models at SS2. Maybe I'd achieve a better result by making a Hi-Res Mesh (MRes4+AppliedDisplacement+Sculpting) and duplicating that... Use Decimate to derive the low-poly model. That means I get all the geometric detail very roughly and then the Normal and AO Maps can do all the work at "bridging the gap" and smoothening it.

These days, I like to weight-paint and rig using Mesh Cages anyway... so maybe a Lo-Res Mesh that's gone through Decimate won't be so much trouble.
Some turbulence here... but still happy. :wip:

[Bao2]

I like the normal one (the middle image). But I would pump some the value (in the map to
panel the "Nor" slider that defaults to 0.5, up it to 1.0 perhaps).

You have some problem with displacement.
Use my image as the displacement, not the one you generated.

The non-manifold mesh is an error probably caused by a welding of many vertices creating an erroneous mesh. Blender can live with it but the script that calculates the bake is not programmed to understand such a mesh. I have created some erroneous meshes myself and blender didn't crashed so I just can't see your mesh (I go to bed and I don't want to have nightmares :p ).

Look at the huge "perspective distortion" of your camera. Place some clones of the dice like the image you posted and try to match the camera. Then a render with same materials like the image. It would be a great exercise try to match 100% the image, cloning the exact number of dice and their positions. And then try to match materials and lighting.

[CGIPadawan]

Hello Jose,

I will probably go into replicating the Dice photograph once the mesh problems are sorted out. I cannot declare the experiment a success if I produce "bad" hi-res meshes that crash.

I must do this again. With a new version of the Hi-Res mesh like I mentioned.
I want the flexibility to bake/displace/sculpt... Then bake /displace/sculpt... until it's finished.

That is not going to work if the model is "Non-Manifold" (error producing model).

Normally, when Blender encounters vertex-over-vertex situations we can see a lot of artifacts in the rendering (the so-called collision artifacting). But in this case there was none.. I never suspected I had a problem because the AO and Normal Maps baked without a hitch!

It was only trying to go into Edit mode of the Hi-Res mesh where I got all these crashes.

So.. it's really come undone on that end.

As for the Displacement Map.... Blender's limitation (as per articles referred to earlier) is that the Mesh "receiving" the Displacement map must have a high amount of subdivision/multires to work easily with Displacement without causing the nasty artifacts I got.

I was just toying with the idea of seeing if an SS-2 model could accept one. But it can't.

However, my latest idea.. is to take a fully-sculpted (non-error-ing) Hi-Res model, duplicate it and use Decimator to create Lo-Res version.

[CGIPadawan]

Blender Equations of Doom:


Crash = (Multi-Res 4 + Subsurf 0 + Displacement Map) * Apply

Crash = (Multi-Res 0 + Subsurf 6 + Displacement Map) * Apply

.......

I think pressing Apply on Displacements at high subdivisions causes crashing....
Now what do I do?

:(

[CGIPadawan]

Wait... I just read this:

http://www.blender.org/development/release-logs/blender-232/displacement-mapping/

Subsurf Meshes. Renderface size is controled with render subsurf level. Displacement really likes smooth normals.
Simple Subsurf meshes. Control renderfaces with render subsurf level. There is a pitfall at sharp edges however if the texture there is not neutral gray.

Yes, I know it's old... but I think the fundamentals still apply.

Okay.. Here's what I try to do.. Use SS3 Applied with Multires 4 Unapplied with the Displacement Map on the Hi-Res Model...

Sculpt....

Bake to Lo-Res Model (which should include the crevices done with Sculpt)..

Then keep the Lo-Res at SS 2.. but push up SS Render Levels to 6?

Any Advanced Blenderite out there thinks this will work? :arteest:

[CGIPadawan]

Okay, it says here: http://wiki.blender.org/index.php/Doc:Manual/Modifiers/Mesh/Subsurf

That Non-Manifold Meshes are the result of one edge belonging to 3 faces or more faces.

I'm pretty sure this is caused by the displacement map!

OR

The Displacment Modiifer!


I may also try this method:
http://www.m-sign.nl/?p=111

http://www.m-sign.nl/wp-content/uploads/disp.PNG

Supposed result from the m-sign tutorial using a Sphere with Procedural Clouds with SS-2:

http://www.m-sign.nl/wp-content/uploads/ballsmooth.jpg

It's so hard looking for workarounds for some of Blender's stranger problems. :banghead:

[Bao2]

I will probably go into replicating the Dice photograph once the mesh problems are sorted out. I cannot declare the experiment a success if I produce "bad" hi-res meshes that crash.

That is not going to work if the model is "Non-Manifold" (error producing model).

Select your mesh, enter edit mode. In the menu of the 3D view there is Select. Inside there is Non-Manifold. And it will show you the problematic areas. Delete that areas because that are non-manifold.

Explanation of non-manifold: Create a plane. Select only one edge and extrude it. Select again the same edge and extrude again in a different direction. Look that edge we extruded: three faces go from it. An edge only must be shared for two faces to the mesh be correct.

But blender is amazing. I tried make it crash with this and blender did a subsurf of this wrong mesh without problems. But of course not all the code is able to manage such wrong mesh.

[Bao2]

It's so hard looking for workarounds for some of Blender's stranger problems. :banghead:

You know your problem: the mesh sucks, has problems.
You know the answer: create the mesh again.

And this time use my method: Create the circles. Join the circles with faces. Done.

Can you see how it takes ages to try to fix a bad mesh instead of creating the mesh again and this time correct? That is a lesson everyone in 3D learnt modeling. It was your time.

Come on, I can't wait to see the render.

[CGIPadawan]

Actually you are correct about Blender being amazing, because I was able to continue producing the AO and Normal Maps even if the whole time the Hi-Res mesh wouldn't survive a simple Select operation.

That is my problem now by-the-way, I can't even select one vertex on that displaced Hi-Res Model without Blender crashing. :(

[CGIPadawan]

You know your problem: the mesh sucks, has problems.
You know the answer: create the mesh again.

And this time use my method: Create the circles. Join the circles with faces. Done.

Can you see how it takes ages to try to fix a bad mesh instead of creating the mesh again and this time correct? That is a lesson everyone in 3D learnt modeling. It was your time.

Come on, I can't wait to see the render.

Well I'll try that too. It's possible the problems have more to do with how I did the Dotted Protrusions (1,2,3,4,5 dot-holes)......I had some faces with tris.....Tris are like Kryptonite to Blender and Triangular faces might have messed up the Multi-Res and SS at higher levels.

I'm quite eager to actually repeat the mesh.. I just want to find the true root cause for the Non-Manifold before I go into that.

[sebastian-koenig]

oh, i think it is not a very good idea to combine multires and subsurf. it will slow or crash blender. and it makes no sense. both are subdividing the mesh at the desired levels, but doing it in a different way. if you want to use multires in order to sculpt or correct the displacement-texture, that's fine, but please don't use subsurf with that. both are using catmull-clark subdivision method, so there really is no point in using both.
and: if you finished sculpting, just set the viewport-levels down to 1. selecting will be no problem then. but rendering will still use your (e.g.) 6 levels. but in order to have the disp-modifier working correctly during bake, you have to "pin" it to the same subdiv-level.

http://img.skitch.com/20090829-8a1ihgfcut1c1dcug49r2ikk1c.preview.jpg (http://skitch.com/sebastiankoenig/b7fmj/multires)
Click for full size (http://skitch.com/sebastiankoenig/b7fmj/multires) - Uploaded with plasq (http://plasq.com)'s Skitch (http://skitch.com)

[CGIPadawan]

Ok,


After a while of fiddling, I think there is nothing wrong with my mesh, only that Blender became unstable at MultiRes4 (MultiRes3 is not a problem).

Also, Displacement Maps do not work well when using abrupt solid black lines.

So with that I have put together a new experiment:

Hi-Resolution Model: Multires and all Modifiers Applied
SubSurf 3
Multires 3
Displacement Map for Skull Face
Smooth (Factor 1, Repeat 15)
Face Count = 1,234,720
http://img214.imageshack.us/img214/4864/hiresss3mr3smooth1x15.jpg


Lo-Resolution Model: No Multires and all Modifiers Unapplied
SubSurf 2 (Render Levels 3)
Baked AO Map on MapTo Col and Spec, Multiply, Col set to 0.5
Baked Normal Map MapTo Nor = 1.0
Baked Displacement Map (Midlevel = 0.5, Strength = 0.22)
Smooth (Factor 1, Repeat 3)
Face Count = 19,424
http://img214.imageshack.us/img214/1576/loresss2rl3smooth1x5.jpg


Think it works now. :)

[CGIPadawan]

oh, i think it is not a very good idea to combine multires and subsurf. it will slow or crash blender. and it makes no sense. both are subdividing the mesh at the desired levels, but doing it in a different way. if you want to use multires in order to sculpt or correct the displacement-texture, that's fine, but please don't use subsurf with that. both are using catmull-clark subdivision method, so there really is no point in using both.
and: if you finished sculpting, just set the viewport-levels down to 1. selecting will be no problem then. but rendering will still use your (e.g.) 6 levels. but in order to have the disp-modifier working correctly during bake, you have to "pin" it to the same subdiv-level.

http://img.skitch.com/20090829-8a1ihgfcut1c1dcug49r2ikk1c.preview.jpg (http://skitch.com/sebastiankoenig/b7fmj/multires)
Click for full size (http://skitch.com/sebastiankoenig/b7fmj/multires) - Uploaded with plasq (http://plasq.com)'s Skitch (http://skitch.com)




Oooooops...

Okay.. I try to apply this lesson tomorrow. Thanks again, Seb. :)

[Bao2]

It looks very well now.
Time to render.

[CGIPadawan]

Hello again,


Once more I must thank my friend Sebastian for giving good advice once again. (If you haven't already, please do go to cmiVFX and plunk your 19.95 right now. You won't regret it. Hehehe).

So... What I did was, copy the Base Mesh for Hi-Res work. Removed the SubSurf Modifier (because it will cause errors with Multires). Added Multi-Res 6 and plunked in Jose's Displacement Map. This time, I was keen to avoid the excessive vertex stretches caused by running the Displacement too deep.

This is the result:
http://img141.imageshack.us/img141/4295/horesmr6beforesculpt.jpg

I then APPLIED the Displacement Map and went into SculptMode in Multires 6 and I tried to clean up the Displacement... as well as add some details of my own.

This is the result at MultiRes 6:
http://img502.imageshack.us/img502/411/hiresmr6sculpt.jpg

I've always wanted to do a Skull with the Shaolin Six Dots! :D

It must be noted that Sculpting is an acquired taste. This was my first time to sculpt and I believe it can take just as long (if not longer) to sculpt and tweak as it did to actually create the Base Model itself! At first, it can seem like a lot of work, but after a while I actually got addicted! :P

One thing to note for first-time Sculptors, the Angle in your viewport directly affects how the brushes work. So you have to keep rotating and "spray here"... rotate and "spray there". It helps to have multiple Camera angles.. and Hemi lamps are good.

More angles follow:

http://img180.imageshack.us/img180/64/hiressculptcam2.jpg

http://img83.imageshack.us/img83/3436/hiressculptcam3.jpg

http://img76.imageshack.us/img76/9257/hiressculptcam4.jpg

I'll be going to the mall in a bit... But when I come back it will be time to BAKE all these details to a Lo-Res Model. Thanks to Seb, I managed to do all this without having to Apply the Multires... and PIN guarantees it will work in BAKING even if Un-Applied!

This means I can even experiment with making a Lo-Res Model using MultiRes instead of Subsurf!

Yes, Jose, we will be starting rendering pretty soon.... finally.... :wip:

[Bao2]

I then APPLIED the Displacement Map and went into SculptMode in Multires 6 and I tried to clean up the Displacement... as well as add some details of my own.


For clarity shake: you applied the displace modifier not the displacement map.


It must be noted that Sculpting is an acquired taste. This was my first time to sculpt and I believe it can take just as long (if not longer) to sculpt and tweak as it did to actually create the Base Model itself!


I miss in Blender some features in sculpt that would do it much better. Imagine you add detail and it is too much, your only way actually is undo what you did or smooth it. It would be great add a tool that deletes what you did like the smooth tool but the difference would be that this smoothing is towards your original mesh so if you continue using it the result would be the original mesh. The smooth tool instead don't stops in the shape of your original mesh before the sculpting but it continues smoothing. If this tool were there like it is in mudbox, the sculpt would be much much much much easy.

[CGIPadawan]

Hello Jose,

Actually in a way, I did apply the Displacement Map by applying the Displacement modifier.
Applying the modifier changes the mesh geometry, such that the Displacement Map would be of no use anymore. You can delete it from the Textures and all that, and the pattern remains.

Here's the result of the SS2 Lo-Res model using baked AO maps, Normal Maps, and a new Baked Displacement Map from the Hi-Res Model:

http://img83.imageshack.us/img83/3950/loresss2rl4.jpg

Will work on painting this now and using possibly node-based materials to get an ivory material going.

[CGIPadawan]

Here's a "first draft" of the finished Lo-Res Dice.

I decided to make it a bit like those Chinese dice where the "1" and "4" dots are red.

http://img23.imageshack.us/img23/4026/0171z.jpg

http://img27.imageshack.us/img27/8518/loresdice1.jpg

http://img403.imageshack.us/img403/6835/loresdice2.jpg

I tried something interesting in coloring this model. The basic Ivory color was taken from the Open Material Repository, and then I painted a Texture Map with the Black and Red Details over a White Space. I then set this Map to multiply over the original Ivory color and this was the result.

I think I can be fairly satisfied with the experiment. This dice definitely does not give away itself as a Low-Resolution model.

I'm sure I can apply what I learned here to more complex models like human actors and such. Thanks to Seb and Jose for their invaluable support in this experiment.

Any feedback is welcome. Maybe I will try to do a pic with many copies of this dice as Jose suggested.

:thumbsup:

[Bao2]

Looks beautiful.

So how is done then? Is it a normal map or a displacement map?

[CGIPadawan]

The answer is: BOTH.

The Normal Map is used to "guide" the lighting, and the Displacement Map just adds that little extra undulation on the surface so that the illusion can be maintained for longer.

In reality, this little project used TWO Displacement Maps. Your original one was used as a Displacement Map for the Hi-Res Model to create the detail in the first place.

This first Displacement is then applied to the Hi-Res Mesh and I used Sculpt Mode to smoothen it in preparation for use with the Lo-Res Mesh. This is basically to create "grey areas" so that the Lo-Res model's chance of success is higher.

What all this time has taught me is that if you force it, the Lo-Res model punishes you with "artifacts" and "stars on edges" problems.

Once the Sculpting result on the Hi-Res is OK...Bake another Displacement Map for the Lo-Res using "Selected to Active" from the Hi to the Lo-res. But in reality, the more important maps for the Lo-res model are the AO and Normal maps. The Normal Map is also created by Baking.

Also, the Baking algorithm isn't perfect and occasionally the Normal Map will have "crosses" and other artifacts... I just fixed those with GIMP's smudge tool.

[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.