Surfaces deformed on curved sculpties

Hi.

I'm having problems with slender curved objects as sculpties.

This seems like a simple request, to make a curved, tapering prim using a sculptie, but I seem unable to get a satisfactory result. They keep having very rough or deformed surfaces although the faces appear quite smooth in the 3D program.

I'm using 3D Studio Max 9 with the SculptGenMax0.4.1 script. I can make it work okay, but I can't seem to get a good result.

I thought I knew about 3D since I've played with Bryce and Poser. But after installing the Max tryout, it's clear I know very little. It's like OMG, there's a million controls. Still, I have been able to follow the tutorials and learned how to work the vertices to get a curve using Soft Selection, etc.

I was making the textures at 128x128 thinking that might help. Then I followed the "nearest neighbor" instructions for resizing in Photoshop and made them 64x64. The result was largely the same. I'm being very careful to move or scale rows of vertices at a time. I don't know where I'm making my mistake.

Can someone offer a suggestion? Am I curving the cylinder the wrong way with Soft Selection. Is Blender easier? For that matter, is it better to learn Max since that's what they use at the JC where I want to take classes? Will knowing Blender help you get a job as much as knowing 3D Studio?

Thank you

http://i208.photobucket.com/albums/bb206/Annalese_album/ugly.jpg

To answer your last question first, it's doubtful that knowing Blender will get you a job anywhere. Blender's strength is that it puts 3D in the hands of the masses, and for that I applaud it, but as is the case with many open source programs, it operates in a very non-standard way. While it's admittedly powerful, you'll find that many professional 3D modelers and animators have a habit of laughing at it, and thinking of it as nothing more than an amateur's toy.

Myself, I wouldn't call Blender a toy per se, but I really can't stand its interface. You asked if it's "easier". Most people would say absolutely not. Because it's so non-standard, Blender can be a tuff program to learn.

3DS Max, on the other hand, is a staple of the 3D industry. Being fluent with it is very much a marketable skill. It's going to have a steep learning curve as well, but there are truckloads worth of books, online tutorials, and (as you mentioned) college classes to help you learn it properly. None of that is true about Blender. So if it's a choice of one or the other, choose Max any day of the week and twice on Sundays.

That doesn't mean you shouldn't learn Blender, by the way. The more tools you can work with, the better a modeler you will be. By all means, learn Max first though. Consider Blender an after thought.

I'd suggest familiarizing yourself with Maya as well as Max, by the way. It probably won't be possible for you to learn every last bell and whistle of both (as you said, there are a million functions), but it's worth being familiar with at least the basics of each. Those two programs constitute 90% of what happens in the 3D industry.

As for your sculptie question, it's hard to say what's going on without seeing your model. Can you post a screenshot of the Max model (shaded with wireframe showing) and a snapshot of the sculptie?

I'm having the same problem, and its not the program you're using, its the way SL is handling the compression and sampling on objects where their vertices are stretched out. Buying a better app is not going to fix this. Max is more than good enough for the job.

the way I find around it is to make the object somewhat differently. Not slender and tapered, but rescale another version of the tapered version to be squat and dense, then resize it in SL using the regular scaling controls to its original scale. That will get you a smoother result.

I don't know the details of the SculptGenMax exporter. But if it is like the one for Wings 3D, it's important to center your model on the origin before you export it.

Another tip, which is probably going to be relevant regardless of the exporter, has to do with shapes like these which are much longer in one dimension than the others. You'll get better accuracy in SL if you make your model be more or less the same size in all dimensions, and then stretch it out to its final shape after it has been imported into SL. The reason for this is that the exporter will probably use the same scaling value for all three dimensions, so a large range of values in one dimension will create more granularity in all three dimensions.

Having said that, though, those errors at the top sure look like compression artifacts. My experience is that pixel-doubling to more than 128x128 does not significantly improve things. But you might give 256x256 a try. Just remember that larger bitmaps will load more slowly, and when one first sees your creation in SL, it may have a distorted shape to start and then clear up after awhile.

Thank you

I didn't save the model, at least with the modifications.

I didn't realize at the time that I could turn off one Edit Mesh and add another and make a bunch of different versions from the same basic cylinder all in one file (I do know though . I just made one, created a texture, deleted the Edit Mesh and added another to modify the basic cylinder.

An example of the prim result is here, although I don't know why the link isn't showing properly: ugly sculptie

Thank you Hypatia, and Omei too I will try your suggestions.

Won't making the scale mostly uniform have the effect of eliminating creative control of long slender curves, or twisting curves for that matter? I've yet to try it, but won't a relatively minor curve on the squat fat cylinder accentuate as x & y are scaled in SL to make it tall and slender?

Per your observation Hypatia, I thought stretching the rows of vertices might have something to do with it and other surface texture attach problems. So, I went back and and did the sculpted textures all over again making sure to keep the 32 rows of vertices as evenly distributed along the length of the cylinder as possible. That image is the result.

Here's what's odd. The first versions produced smoother sculpties, but the surface textures applied with zig zag distortions (attach points?). The second versions of the sculptie textures applied the surface textures much better, but had all of the evident surface problems on the prim.

It seems I created the second problem by trying to correct the first, and it's not the first time I might add

I read someplace here that JPEG distortions are most pronounced near the edges of the image. Do these 3D programs automatically render the texture to the size of the object, or to the 10 unit size? I had the impression it's to the 10 unit dimension, but if it's the latter I wonder if making the model smaller would help, so as to avoid coming too close to the edges?

Thanks again you guys. I'm such a novice with advanced 3D, not to mention these programs in particular, that I simply had no idea if I was overlooking something really basic and obvious.

it could yeah, but you have some control over that if you use the original model to resize as a squat fat cylinder. My experiments have been that a sculpt that is much longer in one direction than another tends to get this problem. What you may want to try doing is to go by my original suggestion and resize the original sculpt. Making it even across the length isnt enough, the vertices need to be denser across the length, which means it can't be so slender as a sculpt. If you really want a slender sculpt - you can also use more than one sculpt.

I think you're also seeing compression artifacts in that sculpt, as Omei said.

Yes, Hypatia, your suggestion to model normally, but before exporting rescale so that the bounding box is approximately a cube, is excellent. (At least, that's how I interpret your suggestion.) But I don't think I understand your second sentence. Can you elaborate?

I just noticed this part of your original post. For mosts purposes, I don't think you should model at 64x64. Model at 32x32 (or lower), create the sculpty bitmap, and then rescale to 128x128 so that each pixel becomes a 4x4 block (or 8x8 or ...) of one color. I can't guarantee that this will give the minimal JPEG artifacts for all sculpties, but I think it will for most.

what happens when you have a stretched out object such as the one here, is that the distances between edges becomes much greater... this needs to be more compact when exporting the sculpt or the sculpt won't be smooth. This translates into a larger color area to sample for that part of the surface, which makes it smoother.

I've done a few tests and here's what I've found, which may or may not apply to Hypatia's suggestion as I'm not sure I completely understand it.

I did some tests beginning with at 10x10 cylinder and putting a bend in it, then scaling down to bring it back within the 10x10 cube dimensions. That doesn't produce an entirely smooth surface, but when you scale x&y in-world it becomes quite smooth. Unfortunately, it loses most all of the curvature too. (Narrowing x&y has the effect of lessening the arc, not increasing it, despite my earlier thought

So, I did some more tests from there lessening the diameter of the cylinder incrementally. I didn't use a standardized method (20% increments with uniform curves), but the results nevertheless seemed to demonstrate that you can reduce the the cylinder's diameter by 75-85% (I'm assuming its a percentage value Max displays) and still get a reasonable result. You can then scale x&y farther in SL which has the effect of smoothing.

Unfortunately, lessening diameter of prim in SL to smooth away surface flaws lessens the arc of the curve too. So, it's a matter of finding a workable diameter in the 3D program, one which allows for more radical S curves, then modifying the result in SL to smooth the surface at the expense of some of the curve. There ome arbitrary, but lopsided ratio of trade offs in other words. ~sighs~

The caveat is surface textures. The wider the diameter of the sculptie texture the less adaptable it is to asymetrical textures, or so it appears. That is, when you bring in something fat and slim it in SL, it seems you're increasingly confined to squared textures. So, a 32x1024 texture won't display properly whereas a 128x128 texture will. I get bad distortions when mapping oblong textures on SL smoothed sculpties.

Would it be possible to post a simple diagram?

I think I understand, although I'm not sure I know how to rescale so that the bounding box is approximately a cube.

In other words, the model stays the same but that bounding box I see around the model, which is now narrow on one axis due to the scaled diameter of the model, needs to be widened so it's a cube again? (I haven't the faintest idea of how to increase the size of the bounding box without increasing the diameter of the model too

OK, I see your reasoning now. My comments were actually addressing a different issue. The full range of floating point values for the vertex positions have to be quantized to only eight bits. Assuminig the exporter doesn't change the shape of the model, that means the distance represented by an increase of 1 in the pixel value is determined by the length in the longest dimension. Hence, extra length in one dimension hurts the spatial resolution in the other dimensions.

At first, I didn't think the quantization to only 8 bits would have a significant effect on the accuracy of the scultpy in SL. But I have found that once I control for JPEG artifacts, quantization errors are the biggest source of errors in my models.

Here's a series of pictures to illustrate. I did this in Wings, so the details will be somewhat different than what you see.

First I made a model that was tall and skinny. See fullview.jpg and closeup.jpg. Then I exported the model. This requires quantizing the floating point numbers that describe the vertices into 8 bit integers, which creates quantization errors. This is a limitation in the sculpty format; no exported can avoid this. To show the effect of this error, I re-inported the sculpty back into Wings. (I chose to do this instead of importing into SL because that would have introduced compression errors as well, and I wanted to illustrate just the quantization errors.) You can see the result of the quantization error in "closeup after quantization.jpg". Really bad, isn't it? The problem is that the wide range of vertex values in the up/down direction has created problems in the other two directions.

To ameliorate this, after modeling and saving, but before exporting, you can shrinkthe vertical dimenstion until the height is about the same as the width. See "full view after shrinking vertically.jpg". This obviously distorts your model, but you can undo that distortion after it is in SL. When I exported this rescaled model, imported it back into Wings and stretched it out to its original height, the top looked as you see it in "closeup after scaling, quantizing and rescaling.jpg". Big difference!

Does that make sense now?

I understand what's happening now, for the most part.

I tried something similar after the last post, using one of the long curvy cylinders and wondering if the model was the only way to make the bounding box a cube again.

Since the arc of the curves I'm using makes X near 10 units and Z stays at or near 10 units tall, I simply scaled the y axis as wide as the others were long or tall and then exported it. I scaled the stretched part back to skinny in SL and the result was much improved.

I'm still having problems getting asymmetrical textures to wrap correctly, but I'm better with textures than I am with models.

Thank you all so much for spending time on this. It helps/helped so much and I really learned a lot