I’m by no means an expert on this, but at least as far as the environment for the animation goes, I’d experiment with pre-calculated radiosity and a buffered spot or two. Assuming the semi-realistic GI type of soft lighting is what you are after.
The dupliverted spotlight technique used to fake global illumination isn’t necessarily that slow either, if you lower the shadow quality enough. In an animation the drop in individual frame quality might not be noticeable.
After spending some tens of minutes calculating the radiosity solution, each frame took an average of 25 seconds to render on an old 1.2G Athlon. The lighting is far away from perfect, but it is only an example.
(Sorry about the large size, I packed it too loosely by accident.)
Indeed, radiosity naturally works best for a static environment, but you could do the background with it and have a different lighting setup for the moving parts. The dupliverted spotlight method could be pretty fast if used only for the moving objects (+maybe the plane on which the objects move, if there is one). Though I still think it’s not that slow even if one uses it for the whole scene.
I applied the textures first, joined all meshes, did radiosity…only had to adjust the painting afterwards.
I’ve not worked a lot with radiosity. I just wanted to say that on my museum project movie, I used 35 lights, raytracing, fields, OSA 16 and Ambient Occlusion. The rendered time per frame was between 4 and 8 minutes…
It is long, but it was a quite realistic rendering.
I didn’t found that the number of lights was the main consumer in computing power.
There is no real answer for what you asking because your question was mutualy exclusive to begin with: I want fast GI but I don’t want to fake. That is not possible in Blender. Maybe with PDI’s renderfarm…
But you can get some nice results using hemi light, sun lights and some clevery placed spots. Otherwise, the technology just isn’t there yet. Just check the render stats from the Pixar movies for example. The normal rendertime is something like 15 hours in every frame where Sullivan is, but on the average it could be a couple of frames a day. I’ve seen some Siggraph papers on fast GI and fast radiosity. But even it is all just relative.
For complex animations, you’re going to save a lot of time by cutting things up. By that, I mean for one shot, you’ll render the background once, with AO or something nice, then render the characters without a background and composite. If you’ve storyboarded your animation, you’ll know what shots you need, and how best to composite them. Your goal is to render as few items as few times as possible.
If you’re making a big outdoor anim, you’ll just end up killing both your computer and yourself if you don’t use compositing. It’s what the pros do.
Keep your lighting to a minimum. The more lights you have, the more calculations per-pixel.
It is a good idea to build up scenes in strips, from back to front. This allows you to render in passes and it also enables the computer to solve the problem in distinct, easy-to-digest pieces.
When a particular light is needed for a particular object, confine the light to that object by using the “layer light” feature. Now the computer knows that it need consider this light only for certain objects, again simplifying the problem. (Such a light will “shine right through” objects that it cannot see, which can often simulate ambient-occlusion quite nicely!)
Sometimes you find yourself using a light only for a particular effect, like “god-beams” shining through mist. In that case, render the god-beams as a separate strip and composite that into the picture.
Focus your attention strictly upon what you are trying to achieve and look for the computationally simplest way to get “close-enough” to that point. Don’t be afraid to “fake it.” In fact, “faking it” is the name of the game. You are under no obligation to “model reality” in your quest to get the picture that you want.
You can do as you say, using backbuffer, whick is usefull in some cases, for rotoscoping. Or use a video editor like Blender Built in or an external one like Adobe Première, Magix, or any Video Editor you like.
Choose one that support a true alpha layer. I’me used to work with Magix, but unfortunately it doesn’t. Blender video editor is a good solution. For my own, I have to learn about it now.
I’m sorry about the minor delay in answering this… Yes, I can provide the blend file if you are still interested. There’s not much to see. There have been problems with my web space provider, so I’d rather email it to you instead, if you wish. PM me if you need the file.
Why don’t you guys use texture baking ? The most simple method for superb effect and short render time.
In blender, select Yafray as renderer, for best result. Set lights as you would do a single render. Go to UV image editor, bake textures. Switch to internal renderer. Attach baked textures, maby as lightmaps if you need more control over textures. Play with some light tweaks.
Render with no light, or one light.