> With scanline and ray-tracing renderers, there comes a specific point at 
> which the renderer is *done* with the image and is ready to render the 
> next image.  Unbiased renderers are never really done; they just go 
> until the user decides the quality is good enough.  Sure, the user may 
> be able to set the renderer to go only a certain amount of time, but 
> except for a very short clip, and only after rendering a couple of 
> frames in that clip, can the time required for each frame be predicted 
> with any meaningful reliability.

On the only unbiased renderer I used, it uses passes. Being *based* on 
random numbers (shooting rays in random directions), of course the time 
each pass takes is non-deterministic, but you surely could render the 
same number of passes for all frames (instead of leaving it the same 
amount of time), and you would get approximately the same amount of 
noise on all.

Post a reply to this message