Cycles Questions

I don’t know enough about cycles yet to know if my questions are possible or not so I’m looking for advice.

When I set cycles to render I am making an educated guess as to how many passes are required to get a relatively clean image. If I set cycles to render with 500 passes, for example, and I find when the rendering is done that more passes are required is it possible to “add” more passes? Is the image held in a buffer that would allow the rendering to continue if required?

Secondly is it possible to save a cycles rendered image at user determined intervals? For example if I want to set a test image to render overnight at 10000 passes is it possible to have Blender/Cycles save a progress rendering at intervals of say 500 passes, so in the morning I could look back over the saved images and see at what point the image was acceptable? That way I would know what settings to use for future renderings of the scene.

Cycles use quasi random sequence, and did not save that information when stop render, so you cannot simple continue rendering to clean noise further. But you can cheat it, you must change “seed” number to get another image and make weighted sum of both (in compositor or other software like GIMP), based on samples used. for example If you use same samples number in original and secondary images, you can just average two images. if samples differ, you must use weight_0=(first_image_samples/(first_image_samples + second_image_samples) and weight_1 = 1.0 - weight0, making final_image = first_image_pixel * weight_0 + second_image_pixel * weight_1. I very doubt anyone will do that, just guess samples better at first time :slight_smile:

There is no easy way to save rendered image at user determined intervals, maybe emulate it using fake animation treating every frame as subframe, and again, you must use different “seed” every time or get equal image that is useless. Typing “#frame” in “seed” enough.

I just finished up a project where I did just what Storm suggested where I rendered to the same sample sized image with different seeds and mixed it together in the compositer. It actually does get some pretty good results.

Also, you might look on BlenderCookie for Bartek Skorupa’s compositing tutorials. He had one back in December or January and a follow up one just last week where he was able to get a very clean final image with fewer samples by compositing the different passes together and adding a bi-lateral blur. I used it on the project I just mentioned, and it looks way better (to me, at least) at 500 samples than a similar image I did back in October with 1000 samples.

Thanks for the advice, I will certainly look into the compositing suggestions. Regarding my second question it occurred to me that I could approach it a different way. Instead of having Blender save progress renders (which it can’t do from your explanations above) I could use a timed screen capture approach. A quick search online revealed that IrfanView has this feature. I’ve just finished trying it out and it does what I need