The problem with 3d

[ghostwheel]

I have a problem with most 3d work for games, movies, illustration, whatever. Ever see a cool piece of 2d concept work from a game or movie and think it looks amazing but the final product looks static, bland and boring? Yeah, me too. I think the process is the biggest problem. 3d (this includes practical/traditional models as well as digital though it's not as severe) is so complex and technical that by the time one gets to the final stage of the creation process, everything that made the original designs loose, energetic and exciting, is refined right out of existence. This is a BIG problem.

Over-refinement is an issue in any art but the process of 3d is especially problematic. I wish there was a better way.

[Chris W]

I think the problem is that 3D art tools are made by engineers, not artists.  When you draw, you are creating a final work.  When you do a 3D model, you are really just making a sort of seed that the simulation "plays".  And the simulation is always mathematically perfect, because it has to be.  I've always wondered what a 3D tool would be like if its goal was not to mimic the physical laws of light, but to be targeted at an end result look or feel - you know, artisty stuff.  I know we've rehashed this before on here, so I won't go too into detail with it.

[God at play]

One of the best solutions I've seen to that problem is voxel-based sculpting. I believe 3D Coat is the first to have that feature.

It might be worth checking out, although the interface is still plenty technical. But once you get used to that, the actual act of voxel sculpting is pretty nice in that regard.

[ghostwheel]

I don't think voxels look very good, especially for what I'm after. I've been learning ZBrush, which alleviates the problem somewhat but not to the extent I'd like.

[God at play]

At a high enough resolution, there's no difference between that and polygonal sculpting. The next/most recent version of ZBrush is supposed to have it.

[Chris W]

I think voxels make the process easier - you don't get ugly artifacts like pinching or stretching and shrinking of volumes, but to the larger point, it is still converted to polys at runtime and rendered using a simulation of the physics of light.  It also creates, by necessity, a perfectly logical structure in 3D space.  But what if we could break out of that rigid logical model? 

Just a random thought I had - what if, instead of pulling verts to construct a logical volume, a tool was set up so you could instead trace silhouettes with your stylus?  You'd start with your front and side views, but it would let you rotate to any angle and trace the appropriate silhouette.  When rendered, you would see the drawn silhouette, or an interpolation between the closest ones available, depending on how many you drew at different angles.  The result would not be a volume that logically exists in 3D space, but there would be enough information that it would hold together visually from any angle, and it would hold the visual life you get from a manually done drawing.

[God at play]

Just a random thought I had - what if, instead of pulling verts to construct a logical volume, a tool was set up so you could instead trace silhouettes with your stylus?  You'd start with your front and side views, but it would let you rotate to any angle and trace the appropriate silhouette.  When rendered, you would see the drawn silhouette, or an interpolation between the closest ones available, depending on how many you drew at different angles.  The result would not be a volume that logically exists in 3D space, but there would be enough information that it would hold together visually from any angle, and it would hold the visual life you get from a manually done drawing.

In computer graphics lingo, this is referred to as an 'imposter,' and is usually rendered as a sprite raster image. I agree it would be cool to have a tool designed for this specific task, especially in vector. Usually how it works in game engines is that they are automatically generated from 3d models.

[ghostwheel]

I think voxels make the process easier - you don't get ugly artifacts like pinching or stretching and shrinking of volumes, but to the larger point, it is still converted to polys at runtime and rendered using a simulation of the physics of light.  It also creates, by necessity, a perfectly logical structure in 3D space.  But what if we could break out of that rigid logical model? 

Just a random thought I had - what if, instead of pulling verts to construct a logical volume, a tool was set up so you could instead trace silhouettes with your stylus?  You'd start with your front and side views, but it would let you rotate to any angle and trace the appropriate silhouette.  When rendered, you would see the drawn silhouette, or an interpolation between the closest ones available, depending on how many you drew at different angles.  The result would not be a volume that logically exists in 3D space, but there would be enough information that it would hold together visually from any angle, and it would hold the visual life you get from a manually done drawing.

ZBrush has such a tool. It's useful for getting a rough outline. It's not something you want for any kind of detail. Also, you don't pull verts in ZBrush, you sculpt. That's why it works well from a artist perspective. But the problem is the final version and the number of steps it takes to get there that lock it into the ridged end result. The complexity is also what makes making 3d assets so time consuming. You can't actually use the final sculpt for animation, either real time or rendered. You have to bake it onto a low-poly model that you have to build after creating the high-detail version. It's kind of crazy.

[Erik Svedäng]

http://rhondaforever.com/

[ghostwheel]

That looks like a fun thing to play with but I don't see how this is relevant to my point. There are some really nice tools like ZBrush. That's not the problem. The problem is the whole process. It's a bit like casting a bronze statue for every character; it's complex, time consuming, expensive and rarely captures the energy of the original sketches.

[God at play]

http://rhondaforever.com/

Whaaaaaaaaaat?! That is amazing! Thank you so, so much for this link

[Erik Svedäng]

Sorry, I didn't try to make a point, I just was too lazy to write anything about it! It has something to do with 3d tools at least =)

[Michaël Samyn]

I agree that the tools we need to work with are barbaric. It's as if nobody consults with an artist before they make an art creation tool. And now we're stuck with methods of defining and rendering 3D worlds that don't even lend themselves well to other interfaces.

A possible solution, or at least a way to work towards more satisfying results, for us has been to reduce the importance of concept art in favor of working in the game engine as much and as quickly as possible. Imagination is your worst enemy sometimes. It's better to start dealing with how you work will be presented as early as possible. And then you work with everything that's at your disposal.

Maybe you don't need to make a perfect mesh if a shader can make it look pretty. Maybe the texture doesn't need to be all that well defined because the physics animation makes things look good anyway. Maybe the character doesn't need to express everything with his face because the sound says more than 1000 pictures. Etcetera.

So: to consider the game engine as the context of your work, and not the modeling tool. This is multimedia!

[axcho]

I agree. It's a similar problem to the problem of programming, though maybe not as bad.

I would love to see a probabilistic rendering system based on something other than triangles, that produces something like this:



It's something I've been thinking about in the back of my mind for a long time, but no grand insights (or experiments) yet. Still, it is a dream of mine.

The closest thing I've seen is maybe the game Love, by Eskil Steenberg.

A possible solution, or at least a way to work towards more satisfying results, for us has been to reduce the importance of concept art in favor of working in the game engine as much and as quickly as possible. Imagination is your worst enemy sometimes. It's better to start dealing with how you work will be presented as early as possible. And then you work with everything that's at your disposal.

I think this is a great partial solution, in the meantime. Kind of like rapid prototyping with game programming instead of just imagining grand design visions in your head.

[Kjell]

I would love to see a probabilistic rendering system based on something other than triangles

Already possible. The primary reason why ( i think ) we don't see this yet, is because you not only need to develop a custom rendering pipeline, but also the ( plugins for ) tools to create content for that pipeline + retrain artists who only have ( years of ) experience with the cookie-cutter polygon based pipeline.

The easier / more cost-efficient alternative is to use a post-processing effect that mimics some of the aesthetics you're after.