I’m trying to design a piece to be made at shapeways. It’s a flat rectangular object with an eagle and stars.
I have a grayscale image for displacement. I use Multires, subsurf, displace, and smooth, in that order, but I can’t get the face count high enough to make the eagle look good.
Actually, I can, but then I can’t do anything with it. It takes about 12 million faces for a single displaced grid to make it look good. The final face count needs to be under a million for shapeways to accept it, but blender chokes when I try to use decimate to reduce facecount.
I tried doing a lower res (3 million faces) version, but it still hangs on decimate. My theory was that I could decimate a low res mesh, and then re-apply the displace map to get more faces where I need them. (much of the mesh is flat and smooth, and so doesn’t need a high face count)
I have a 3Ghz Athlon II 250 (regor) dual core chip, and 5 gigs of ddr3 ram. My graphics card is a (brand new and recently installed) Radeon 5670.
Do I need more computer to handle this job, or am I just doing it wrong?
It’s highly probable that Blender does not in fact hang, it just needs some time to complete the job. Since the algorithm is single threaded and takes exponentially more time to complete for more complex meshes, try to just wait it out. Give it an hour and only then give up. Ignore the fact that the OS is telling you the application isn’t responding or whatnot.
If that fails, try using MeshLab for the decimation instead.
What is the topology of the object you’re trying to displace? If you cut edges along the lines in the picture the displacement will look good with much less subdivision.
Also, if you think the displacemet is too sharp you can add another subsurf after the displace modifier.
And yes, your cpu is a little too weak when it comes to handling high polycounts.
Waiting an hour is standard practice for me ATM. I let Meshlab run overnight, last weekend, in attempt to use Quadratic Edge Collapse. In that case, Meshlab wasn’t even able to open it. I made a smaller facecount version, and managed to open it in meshlab, but canceled the decimation after about 2 hours.
I watch my performance monitor while things are running, and Blender never uses more than 50% of combined CPU. The vast majority of which is on CPU1.
Also, I made a file yesterday that was 1.21 JiggaBytes!
Displace on a 8.5 X 1.7 X 0.125 “cube”. Restricted to Z axis so I get a raised image on the front, and indented image on the back. This is my preferred method, because it has 2 useable surfaces for molding rubber on, a positive and a negative.
The second version is just a displaced grid with the solidify modifier used to give it thickness. This results in a smaller file, but I still can’t get the detail I need using this method.
12 million faces, good grief! That’s a little too many faces! You are over heating your CPU big time!
Option is clear. You either wait 10 years for the PC to become more powerful or use normal mapping. What is your objective here? Do you want to make the eagle really looking 3D, or add just some relief texture to it?
If you want 3D, you need to model it by hand; which is not all that difficult. It will be little time consuming; just make rough outlines using faces and subsurface it.
By example if you make a normal map in GIMP with the normal map plugin you can obtain with only 1 face what is show in the following animated gif (click to see) :
But as you want to make a model for shapeways, unfortunately normal maps will be useless in that case as shapeways needs real geometry.
Maybe modelling manually as ridix mentionned, would be then the best solution using the picture as the guide with the background image function, as Blender (at least on windows as people mention this is not true for Blender on linux due to much better opengl on those OS) is not good with millions of faces like you want.
I’m reading up on selecting and subdividing faces to get better use of my facecount. I’ll see how far that gets me before I resort to manual modeling. I have quite a few images I want made to relief work for molding.
I’ve got it working much better now. I used an array of cubes, which I flattened. I applied the subsurf and displace modifiers, and then traced around the displaced areas of the mesh with grease pencil.
I then went into edit mode and subdivided the mesh, switched to wireframe, and used Ctrl-LMB to trace around the grease pencil, subdivided that selection, and re-traced the grease pencil, and then subdivided the selected faces a couple more times.
I went back into object mode, reduced the subsurf level, and applied the modifiers.
Exported to STL, decimated in meshlab.
Looks pretty good now at just under a million faces!
Could we please compare results - i took your image, set resolution in gimp 600x600 ppi, applied Gaussian blur 2x2 and used BW version to displace (color and spec too) plane. No UVs, just Generated. In order to get squares, i subdivided plane’s lenght by 3, then all by 300 or 400 getting as a result 1M range. Second image is eagles head, approx same resolution, UVed to show displacement resolution. As i see it is as close as it could be to the image. Gaussian blur on image helps smoothing outlines. Handles pretty easy on PC slightly worse than yours.
If you need reverse, just set minus same displacement or render another plane rotated 180. Pict: Displaced , UV part
Oh, yeah… Guess i’m using bump, since i don’t see displaced mesh in 3d view… Gonna try with modifier. If that works like texture all should be fine i hope.
I took empty with image on it, scaled up image approx 5 times, took plane, resized to match image. Then as i wrote - lenght ctrl-R 3, giving me figure made of 3 squares, after that subdivide all 300 to 400. Result is near to square mesh.
Are you using same .jpg which you posted here? It’s a bit messy. (This is what i came up with). Make it pure BW in gimp or PS; grey levels make bumps, so if you shift with brightness levels you’ll get different heights for pixels.
Edit: Works as expected, still picture needs to be precise as to the grey levels. There is brightness and contrast adjustable if picture is used as texture, don’t know if that affects displacement modifier tho.
Again, if you need cube, select all vertices and extrude z some. It will bring this to some 2M verts. For me it crashed while rendering.
I think I get what you said about resizing and loop cutting the plane. Still a little fuzzy about how to put an image on an empty, and what you did with that. I’ll have to read up more on empties.
I see what you’re saying about extrude. That seems like a much better solution than solidify for a flat object.
Btw, how are you getting that yellow/green color? Is that a material? I haven’t gotten into materials yet since I’m mainly only concerned about geometry atm.
For that empty stuff - i was just eyeballing image proportions to get mesh plane about right. I think that helps Generated mapping be correct. For empties - there’s availability to set pics on them in Properties panel - select empty and check at where you look for say, lamp properties…
Since pic’s proportions are near 1/3 and subdivide does it’s job for every segment you do loopcut length in 3 parts at first.
Green is set as base material, then add same pic texture in a new slot and set it to affect Specular (slide down material’s Specularity before) . It will show it in that Blender Pretty Pink one finds at Texture panel end… Just change that pink to yellow or whatever, maybe Mode too, depends.
Quick, dirty, still better seen onscreen.
Have you considered using vector image, .svg, Inkscape can Trace bitmap outlines, allows curve-edit them, make nice consistent gradients?
And while straight such .svg import to Blender brings some problems, from Inkscape exported bitmaps are much cleaner. @Witold_Jaworski how to trace bitmaps.
