|
|
| Post Number: 11
|
carnage 
Group: Member *
Posts: 435
Joined: Apr. 2007
|
 |
Posted on: Sep. 05 2007,20:20 |
|
   |
Today iv had a bit of success in improving the procedural generation of cracked rock heightmaps. I have been playing with blend modes and tweaked the settings to give a much nicer shape to the groves and cracks in the rock. Also i have managed to simplify the process in just one layer that makes it very easy to work with
http://img160.imageshack.us/img160/9008/rockgenerationek1.jpg
the one layer is simply transparent with black lines that represent the cracks the blend settings then turn the entire thing into a detailed heightmap. The advantage of this is that by simply paining black lines with a brush you can instantly draw your own cracks
i found this useful for adding some smaller detail. The crack inside the white circle was created by hand (not using the difference clouds to generate it). So its possible to easily paint on a extra layer of detail
|
 |
|
|
| Post Number: 12
|
ratty redemption
"understood..."
Group: Member *
Posts: 1351
Joined: Sep. 2006
|
|
Posted on: Sep. 05 2007,22:37 |
|
 |
very interesting and that certainly sounds a lot quicker then how I've been modeling the cracks in my base mesh.
so far today I've not yet found time to do any tests with the heightmap you gave me this morning but I could now try your latest one if you want?
we seem to have two main options once the heightmap has been created. either use it to displace a mesh in blender then export it into d3's renderbumpflat, or use crazybump to render a normalmap straight from the heightmap.
we know that donkey and kat have been getting very good results with the latter, and I found with smaller cracks and other details the blender meshes need a very high subsurf (sub division) level. so maybe the only reason for using blender would be to tweak the mesh or use blender's procedural textures to add in more detail, not contained in the heightmap.
does that all make sense?
|
|
|
|
| Post Number: 13
|
dONKEY
"whatever.."
Group: Member *
Posts: 121
Joined: May 2006
|
|
Posted on: Sep. 05 2007,23:00 |
|
|
I *think* both approaches to normal map creation are valid. CB is such a good application now (actually it probably always was, and it's my use of it that has improved). The thing that to mind makes working from source images the better route is the time saved. Closely linked to this is the ease in which 3d height variation represented by the 2d image is achieved using a normal map generated by CB from a phot source. Achieving this from the high poly mesh approach is of course possible, and in many ways would vouchsafe a far greater degree of control. The big draw back though is having to hand create those same height variation realistically and within an acceptable time. My gut feeling is that depending on exactly *what* the normal is for will dictate the approach that is most appropriate.
|
|
|
|
| Post Number: 14
|
ratty redemption
"understood..."
Group: Member *
Posts: 1351
Joined: Sep. 2006
|
|
Posted on: Sep. 05 2007,23:35 |
|
|
interesting, thanks donkey 
|
|
|
|
| Post Number: 15
|
carnage
Group: Member *
Posts: 435
Joined: Apr. 2007
|
|
Posted on: Sep. 06 2007,00:31 |
|
|
yes interesting, looks like ill have to download crazybump and see what i can do with some of these heightmaps. perhaps it will allow for a higher quality normal map than i am getting from the nvidia plug in that is a bit limited in its settings
working from photo sources i a quick way to work but does not always lead to the higher quality results and is limited to what images you can get your hands on to play wit. even if you have a digital camera its not a 5 minute job to travel to a site with a lot of rocks and take the needed pictures
converting the hightmap into a blender model was what i had in mind when i saw this thread ratty but you are much more experienced at working with things like this than my. I also thing the problem you are getting when you shine a torch directly onto the rock is caused by your sub surf modifier causing the cracks to smooth out too much at the bottom. perhaps creating a nice seam along the crack lines might do the trick
it will be interesting to see what further can be done to this in blender because i cant see how i can improve it much further in photoshop
ill send you the latest hightmap and i myself will see what i can do with crazybump as i want to try add another layer to make the surfaces of the rock sections less smooth
|
|
|
|
| Post Number: 16
|
ratty redemption
"understood..."
Group: Member *
Posts: 1351
Joined: Sep. 2006
|
|
Posted on: Sep. 06 2007,01:54 |
|
|
(carnage @ Sep. 06 2007,00:31)
QUOTE yes interesting, looks like ill have to download crazybump and see what i can do with some of these heightmaps. perhaps it will allow for a higher quality normal map than i am getting from the nvidia plug in that is a bit limited in its settings
cb (crazybump) is very power but I personally dislike the interface, except for the d3 preview window, that is very cool.
QUOTE working from photo sources i a quick way to work but does not always lead to the higher quality results and is limited to what images you can get your hands on to play wit. even if you have a digital camera its not a 5 minute job to travel to a site with a lot of rocks and take the needed pictures
agreed and same apply if we're searching for suitable source photos on the web. I've spent days in the past looking for some.
QUOTE converting the hightmap into a blender model was what i had in mind when i saw this thread ratty but you are much more experienced at working with things like this than me.
understood and thanks but I'm still relatively new to normal and height map creation. I would say they are an art form in themselves, especially as some of the subtle differences in the source images or models can make a big difference when their rendered.
QUOTE I also thing the problem you are getting when you shine a torch directly onto the rock is caused by your sub surf modifier causing the cracks to smooth out too much at the bottom. perhaps creating a nice seam along the crack lines might do the trick
thanks and thats what I first thought until I removed the bottom edges of my cracks. also if you look closely at this shot, your see none of the rock pieces actually touch.
after renderbumpflat has rendered the normalmap, all the cracks and crevasses have been filled in as shown in this next shot. it seems that renderbumpflat extends the outlines of the mesh and then blends any nearby borders together.
QUOTE it will be interesting to see what further can be done to this in blender because i cant see how i can improve it much further in photoshop
ill send you the latest hightmap and i myself will see what i can do with crazybump as i want to try add another layer to make the surfaces of the rock sections less smooth
so far I think you've done a good job in ps, but there could be other filters your not aware of or haven't thought of trying yet. I've outgrown psp7 so need to try gimp asap because I currently can't afford ps, although I might buy it one day if gimp isn't powerful enough.
Edited by ratty redemption on Sep. 06 2007,01:57
|
|
|
|
| Post Number: 17
|
ratty redemption
"understood..."
Group: Member *
Posts: 1351
Joined: Sep. 2006
|
|
Posted on: Sep. 06 2007,03:07 |
|
|
I've not yet tried tilting any of the rocks as I intend to do next, but I have finished filling in the large gaps. I also need to add some more internal cracks into some of the rocks, although I'll probably model those after the tilting.
anyway, here's a few shots of this wip.
notes: the square white wires near the middle of the shot, indicate where the texture will be cropped in psp, with the yellow highlighted mesh object creating the tiling border.
here is that object on its own, drawn in blender with my uv test patten showing which of the border rocks have been duplicated.
notes: the outer square yellow wires and the small textured squares in the corners of those wires. these are so I could quickly do a uv face select mode > u key > project from view (bounds) as I was modeling.
heres the resulting low poly test render from d3's renderbumpflat.
edit: on our pc here renderbumpflat seems to take the same amount of time to render either the low poly or the high poly models, but although goofos's .ase script is very fast at exporting, it still takes a minute or two to export the high poly version, which is one of the reasons I often tested just the low poly as that exported in just a few seconds.
notes: the slightly darker, more saturated outer edges of the image. I think this is because renderbumpflat didn't find any nearby edges to blend together.
although in the next tiling test shot, those darker edges have been cropped in psp as part of the tiling process.
Edited by ratty redemption on Sep. 06 2007,15:16
|
|
|
|
| Post Number: 18
|
ratty redemption
"understood..."
Group: Member *
Posts: 1351
Joined: Sep. 2006
|
|
Posted on: Sep. 06 2007,15:09 |
|
|
these test shots are using carnage's heightmap originally created in ps.
first is the displaced mesh in blender.
notes: the mesh was a subdivided plane and has just over 262,000 polys, which almost chokes this pc here although that might be due to it only having 512mb ram.
the mesh was given various modifiers (mainly displace, but in this case no smooth modifiers) to add in some high frequency detail, but those combined with the high poly count took nearly 4 minutes to export from blender, which was a pain to wait for.
that is why I also use a simple subsurf modifier which when turned off, lowers the mesh to just over 65,500 and that enables me to work on and test export the model more easily in blender.
here is the renderbumpflat results.
and the tiled test shot from psp.
notes: the tiling is slightly broken, I think by the displace modifiers but that could probably be fixed in blender or an image editing program like ps, psp or gimp.
these next shots used the same save game from my previous d3 tests, so the lighting and camera angles are the same.
the first shot uses just the normal map.
and this one combines my previous d3 _h.tga heightmap to add in even higher frequency detail. I also tested carnage's heightmap in the d3 material shader but the results were almost identical to not using any d3 heightmap.
and finally for now, my diffuse and specular textures were turned on in the shader.
|
|
|
|
| Post Number: 19
|
kat
Admin
Group: Admin
Posts: 1325
Joined: May 2006
|
|
Posted on: Sep. 06 2007,17:28 |
|
|
Just a general comment on what's been said so far (implied and actual). As donkey mentioned above, I don't think this is an issue of the validity of the results as both methods are capable of producing good normal maps regardless. From that point of view it becomes a 'skill' thing rather than a 'capability' (are you 'able' to rather than 'can' you), the main issue at hand really is time, secondary to that is the available of horsepower horsepower (for highpoly meshing).
Although id software were a bit 'anal' about doing normal maps from highres models, i do understand that in the wider context of the content creation process within the industry, they are an exception to this and not the rule, certainly now that people have got used to the process overall.
Back on topic : Ratty, what I'd suggest doing at some point is trying to sculpt something like this.. right now you know you can 'polymodel' the rock work but you might find being able to sculpt quicker as well as being able to get more 'organic' results - what you have right now is good but it still has that 'constructed' feel to it primarily because you've not yet addressed the 'randomness' you need to introduce to make the surface overall more convincing (i.e. so it doesn't appear the same depth/height across it's surface [as mentioned above]). Mind you against that is being able to 'tile' the resulting surface.. haven't quite figure that bit out myself yet (sculpting and polymodelling for me is too slow with CrazyBump at my disposal).
Keep going with this though as the one you are doing for certain is increasing your understand of and capabilities with using Blender. Blimey, I only meant to write a few short sentences 
--------------
Admin
|
|
|
|
| Post Number: 20
|
ratty redemption
"understood..."
Group: Member *
Posts: 1351
Joined: Sep. 2006
|
|
Posted on: Sep. 06 2007,18:03 |
|
|
hehe, thanks kat and I think of the recent tests I've done above I prefer the control the poly modeling gave me, although it was a very slow process but it was good practice at low poly modeling in blender. if I liked crazybump's interface I might use it more then just doing occasional tests.
tonight I plan on adding in some variance to the base model which I think will be reasonably quick and probably quite fun.
I've also been thinking about doing more sculpt tests, although on small but highly subdivided mesh sections, then use psp or crazy bump to merge them together to form the 1024x1024 normal map. that way I might get the level of detail I would want for the cracks without choking our pc here.
|
|
|
|
|
|
|
