Rendering Larger Scale Scenes

HI guys,

This is more of a question type thread than anything. I am slowly moving away from the fledgling digital artisits’ tutorial level cup and table top scenes to more complex works of my own design and I am finding a disturbing trend.

I have noticed that as my scenes have been becoming more and more complex that I am having trouble rendering them. I am certain that it is not an issue with my PC but I think with the way I am usuing Blender. It seems that my larger file size scenes (say over 500 MB) are causing Cycles render engine to crash out when it is going through the building objects phase of the render.

I was thinking that it might be wise to start setting up render layers and separate some of the objects onto other layers. is there any good reference on rendering larger scale scenes that is worth reading? Or any advice from the guru’s out there? any direction would be greatly appreciated!

(^-^)/

Compile Blender yourself, and you get much more useful info about crash. Blender must not crash in any situation, otherwise it is BUG. If it ask for memory fore example, and OS can not resolve it, Blender must show error box with something like “not enough memory to allocate temporal image cache”.

I have never tried compiling it myself. I usually just run it off of the installers provided. I will have to look into how to do that.

Otherwise it is not showing any memory errors. Which is why I am not thinking that it is an issue with the PCs limitations. It is getting to the point of building the objects within Cycles and then, POW, it stops responding the screen goes white and windows sends an error saying that it has stopped for some unknown reson please restart your application. So I decrease the file size by deleting objects or what have you and it seems to function fine. just when it hits that magic 500 MB mark it goes belly up. Go figure.

I will look into that compiling idea though maybe that will clear it up.

You say you are rendering with Cycles. If you are using GPU, you need to ensure your card has sufficient memory. 500 or 1024MB will not go far if you are rendering complex scenes.

Moved from “General Forums > Blender and CG Discussions” to “Support > Lighting and Rendering”

Also remember that “big scenes” in a movie are usually just establishing-shots or passing-shots between one major section and another. They are a typical target for “the Ken Burns Effect.” Most scenes in a movie, after the establishing shot, follow a MS / CU / ECU+OTS sequence that takes you to a particular part of the overall set, and then conveys the action in a series of cross-cuts. Furthermore, each of these sets, although they look consistent with the big-shot, are in fact separately built for the purposes of the shot. When the show moves to the next set, the shots will briefly step out to a common point-of-view and just as quickly step in to the new location. Even though the location, as shown in the establishing shot, might be large, each shot is basically just as tight as it can get. (Six inches to the right of that rain-barrel in the dusty Western town is … a plastic chair where the sound guy’s sitting with a microphone, just off-camera.) Look at sets on some studio tour and you’ll be surprised at just how minimal they are.

A “big scene” does not have to be highly-detailed and it shouldn’t be. You’re only showing the lay of the land.

Shoot big-scene, then put some action in one place, then CUT TO a MS that shows the viewer, very promptly, what that action that “caught their eye” is. (Their curiosity must be satisfied at once.) Presto… the story is under way.

If you like, make a blend of the “big set” and then make various shot-specific groups that you can link to. Incorporate into each shot-file only the geometry that is actually needed for that shot. Freely mix Cycles with BI rendering techniques, with compositing, so that you can get the shot done as quickly as possible.

The human eye, having seen something once, and absent any other visual indication to the contrary, will generally see what it already remembers seeing.

When working with large files don’t forget that linking and instancing is your friend, and study what they do on the open movie projects. If you are already familiar with these, then they should help someone else.
http://www.sintel.org/news/library-linking-asset-management/
http://mango.blender.org/artwork/env-art-videos-roundup/

Sundial Makes a good point, Model for your Camera POV, rather than trying to model extreme detail throughout the entire scene, where it might not be seen.

On a related note, is there a way to have a high poly asset with UV textures and decimate it or reduce polys for use in the background, but still maintain the textures?

Thanks Guys,

These are really great points to help out!

Sundail, Thanks for pointing out the camera POV point. I think I will need to do some reworking of the scene to be dure that I am not over doing it. Also I think the compositing different elements into the shot instead of trying to render everything in one go is going to be key. I am going to try another pass at this and then get back with a posting to everybody with how their ideas helped.

Awesome! thanks everyone!

Specifically, what I suggest is this: start by “blocking out” your world … and by “blocking out” I literally mean, blocks. Simple shapes, the roughest modeling you can manage, but “to scale.” In other words, when you superimpose a long-shot from your finished film with the blocks shot, in an ideal world it would almost match.

Now, get down there into that world and start imagining your movie. Insofar as long shots are concerned, put the major shots “on the map.” Then, start with also-blocks rough sets for the closer shots, also to-scale. Always include a human-sized shape in the shot. (Sphere, cylinder, two cylinders for legs, two for arms. Done.)

What you’re doing is working out what each shot will be, and that determines where you do and do not need to be concerned with modeling of any sort. If you do not determine that you need it, it doesn’t get built. Lavish attention based on how much is or isn’t seen. If you can’t see it, don’t decorate it. If the detail won’t affect a substantial number of pixels, don’t add the detail.

When you get to filming individual scenes, you don’t know in advance what’s going to be the final makeup of cuts. So set up different fixed camera-positions, including dolly tracks and booms and so-forth, and shoot animatic-style “film.” Then edit that. This guides your refinements. This tells you what frames you need. This locks in what you can and can’t see. You make very fine decisions about f-stop, camera angle, moves, lighting and so-forth using OpenGL renders and to-scale geometric shapes. You are going to lock in those decisions; you are going to make them now. You hold up that geometric mock-up against the final and it matches.

You know that the time-waster on any and every CG effort is render time. So you absolutely do everything that you can, first, to minimize that; then, to do it in such a way that whatever you do render can somehow be kept and used.

Real-life films are (sort of…) “shoot, then edit.” CG is “edit, then shoot.”

1 Like

That is infinitely quoteable!

Thanks! (^-^)/