What is the aim of a highpoly for inorganic models?
I'm curious here.
Everyone is doing highpoly+lowpoly and baking normals from one to the other right?
I can easily understand the need for highpoly sculpts for organics because there's simply too much detail to be had, so you sculpt it and transfer it in the normal map.
But for hard surface stuff, and inorganics, what is the fundamental purpose of the highpoly version? Is it to provide more surface detail? Or to help sculpt the overall shape? Or to simply bake a nice looking AO map?
I'm asking so I know the main function of the process. From all the tutorials I've watched nobody seems to have answered that question, rather they just go and bake without really saying what they would bake and why.
For instance, when would you choose to bake something and when would you model it in the low poly? What shapes would be better with either approach? Can you give examples of what shapes are better off modelled in a highpoly and how to do those?
I know, big question I guess. But would be helpful!
Thanks.
Everyone is doing highpoly+lowpoly and baking normals from one to the other right?
I can easily understand the need for highpoly sculpts for organics because there's simply too much detail to be had, so you sculpt it and transfer it in the normal map.
But for hard surface stuff, and inorganics, what is the fundamental purpose of the highpoly version? Is it to provide more surface detail? Or to help sculpt the overall shape? Or to simply bake a nice looking AO map?
I'm asking so I know the main function of the process. From all the tutorials I've watched nobody seems to have answered that question, rather they just go and bake without really saying what they would bake and why.
For instance, when would you choose to bake something and when would you model it in the low poly? What shapes would be better with either approach? Can you give examples of what shapes are better off modelled in a highpoly and how to do those?
I know, big question I guess. But would be helpful!
Thanks.
Replies
That amount depends greatly on the prop - the bigger the scale of it, the larger that amount would probably be; the more important a certain detail is, the less that amount would probably be.
Most other things can be left to exist only in the normal map as they could be considered surface detail rather than structural form.
i.e. The low poly should capture the silhouette of the shape, the normal map captures the details on it.
The same principal applies to organic and inorganic models - it's not specific to one or the other.
A good way to think about it is imagining there was only a diffuse texture and asking yourself "If this detail was only in the texture and not the geometry, would it look weird?"
For even a car, there's a lot. Think of all the vents and the mesh grills. the headlight and taillight 'jewelery'. The part numbers (if you're polyphony):
http://media.photobucket.com/image/GT5%20Prologue%20BMW%20headlight/Sylon03/Gaming/GT5/EigerNordwandShortTrack.jpg
If you're doing a tiled surface, you make some tiles, a grout layer, Then grab and rotate random tiles by 1 degree, and some tiles you grab the middle and make it a tiny bit concave to make it seem more realisticly imperfect.
Tyres are a nice example. You can either model them from nothing, of make a heightmap, distort the resulting mesh, then bake. Look at the detail on your car's mudflaps next time you drive (and if you have them?).
A normal map replaces the mesh normals of your lowpoly model with that of the highpoly model per pixel. What this means is you get much more accurate lighting on your hard surface models.
So you can't just think in terms of small details like screws and bolts, its the entire thing. You're replacing that nasty vertex based shading with high per-pexil based shading. When you bake from a high res source, you also end up with a seamless result that, in theory, is going to look like the highpoly mesh(this is the goal at least).
When you bake, the renderer also compensates for the shading of the lowpoly, so it isn't something you can paint by hand either. You need a quality, accurate highpoly mesh for best results.
Now, as far as how and what to model in the low, it comes down to:
A. What detail is important/most likely to be viewed.
B. How close you will get to the model
C. How large the model is
D. What technical restraints you have(texture size, poly limits)
There isn't an always do this and never do that sort of rule. Generally you'll model the low and medium frequency details, and leave the high frequency details for the normal map.
Earthquake: so I guess my problem is that the normal map isnt a displacement map so you would always have to model out anything that can be seen in silhouette at least right? So is it safe to think of a normal map as adding to the surface details (roundness, depth or whatever) for surfaces where they are generally seen from approximate to the normal (i.e. face on) rather than at a perpendicular (side on)?
Trying to get a feel for how everyone figures out what to actually model and what to fake with normals. I know the answer is a lot more vague, but at least some kind of ballpark would be useful.
Anyway, thanks for all your answers. Appreciate the info.
search around for some hard surface work and you'll see where they're using them. EQ has even posted a lot of work that demonstrates this.
Anything that doesn't effect the silhouette can be included in the high poly/normal map. Panel seams, hatches, raised or indented lettering.
You can also apply maps to your high poly, such as scratches or dents and they will be baked also. I've used a scratch tiling material masked out by vertex pain. then painted the scratches in various places, like on the edges of things or on specific places that would be worn down.
I think Ghostscape was talking about threads like this?
http://www.polycount.com/forum/showthread.php?t=85669
It really comes down to mountains of practice.
The best way to figure it out is to see what does and doesn't work. There are so many variables across all of the different types of models that can be made that has to be balanced with the games priorities and the tech its running on. There is so much that factors into any given decision, that you just have to jump in and work out solutions.
The more experience you have finding out what does and doesn't work, is the type of person that companies find valuable. They aren't looking for someone who knows all the little rules, but someone who can weigh out the pros and cons of various methods and create rules if needed, but they are probably specific to that company, that tech, that game, that model.
Personally, I do not like to do this, for a few reasons:
A. Its super easy and fast to do this in photoshop, or 3d paint a mask in mudbox etc.
B. Your highres needs to have some form of uvs
C. You dont get an easy to use mask for the diffuse and specular areas, as you would with standard techniques
D. Much less flexible, once its baked down thats it, if you need to tweak it, you have to rebake.
E. Resolution issues, what if your scratches were too fine for the res of your low texture? Rebake again...
So yeah, IMO doing this is more difficult, takes more time and is less flexible.
[edit] Mark: your link goes to this thread
Right, so you only use the highpoly as a target for the normal map and AO baking? I think that makes sense to me (because I hate having to fiddle with UV's too much anyway).
I'll have a look at the highpoly threads again. Thanks all.
B. It's really not that hard. Planar maps or auto unwraps based on smoothing or angle will cover about 90%. The remaining hard to unwrap stuff, can be done with point to point seams, and peel in max2012. They don't have to be fully optimized and prefect and the map size can be gigantic, the UV's just need to be good enough to get the job done.
C. You don't have to use vertex paint you can use any masking method you want. Including methods like Edge Finder. You can just as easily paint a pixel based mask, on the model using Viewport Canvas. You can bake that mask out and use it on your low poly also, just stick it in the diffuse slot of the high poly and bake a diffuse map.
I take my spec maps into max and use viewport canvas to punch up different areas by painting directly on the spec map, as it is on the model. Because it paints flawlessly over seams, they aren't a bother like they are in photoshop.
D. That's just par for the course, if its not rebaked for something like this, it gets rebaked because the smoothing changed or you make a geometry edit. Rarely do you ever bake something once heh.
Opps meant this link: http://www.polycount.com/forum/showthread.php?t=85669
It wont take me all day to do a basic uv on my high, but it will take time, time better spent using more straight forward methods. When you couple this with the fact that you're almost always going to be doing some bump -> normal work when you get to the texturing phase, its just a waste of time to do a small portion of it directly on the high.
A. Just paint it in 3d, in mudbox then, directly on the low. Solves your seams issue.
B. Still work, still time, its not free
C. Yeah, I certainly wouldn't bother with vertex maps, again, just 3d paint directly on the low. Edgefind etc? Now you're inventing excessively complex solutions to your excessively complex workflow.
D. I do multiple test bakes during the lowpoly creation, once the low is approved, I bake it once and forget about it. Unless there is a change request after the low is approved(very rare). If you're doing multiple bakes after your low is finished, you've got some holes/kinks in your workflow.
Yes, for normals, AO, and additionally a color mask. I apply basic diffuse colors like RGB, or RGBCMYW to materials on specific polies, this lets me easily extra layer masks in photoshop for different material types(does not require uvs in most apps).
B. Sure you need to evaluate the time it will take vs the pay off and in some cases it might not be worth it to unwrap the high but there are times that it will. I wouldn't turn my nose up at the method and say don't ever use it. Tools in the tool bag =P
D. Well its not hard to do 2 separate bakes (bump details & High poly geometry), you just turn off the bump on the high poly material. Then you can combine them in photoshop as you like, if you're going to monkey with your maps.
Personally (and I think you taught me this) its better to have it all set up and bake it out and then not mess with it, if you do end up having to rebake you have to go through all those tweaks to get to the end result again. If its all on the model you just bake and you're done.
D. RE: double bake Yeah, but again, then you just have a baked down map, if you do the work in photoshop, you have non destructible layers. If you're just baking out a greyscale mask, then thats essentially the same thing as painting a mask in mudbox or whatever, just that baking it likely takes more time. However, you're not dealing with a locked normal map that needs to be rebaked, so that is a much better solution. You do have to rebake instead of simply resaving a TGA though, and IMO that is more time + more hassle.
RE: What I taught you(keke): Right, this is in reference to those weirdos who will separate each mesh chunk and bake it separately, then combine. Or people who will bake 2-3 maps at different settings and mask them together. This stuff is a mess.
I do many, quick and low resolution test bakes while modeling, to check for a variety of problems like ray trace skewing, smoothing errors, uv distortion, resolution issues, etc. But when my low is done, I bake that shit and forget it. Sometimes I will forget 1 thing and make a minor change and rebake. I virtually never rebake after the texturing process has started.
B. Its the whole "oh we'll just texture the highpoly" pie in the sky idea that a lot of people had back when we started using high res geometry, its not actually any faster, and in may cases, much less flexible. There is a reason it hasn't really caught on that well.