I was also doing render tests at this stage, to try and keep a lot of parameters, such as samples at low levels, to maintain a decent render time. I also had to be careful with assumptions about the final look, especially in relation to the hair.
Here is the final view port wire and shaded mode (Fig.16 & Fig.17).
This was my first project ever that I rendered for so long. The hour count reached six weeks overall. The rendering size was 3600x4500. This was an experimental process for me and I was prepared for this long process from the beginning. I would never ever wait that long again though. The main reason for this crazy length was because of the hairs which were all geometry this time, fully responding to GI, and ray traced shadows cast by transparent textured hoods on them. Also almost every element in the scene had glossy reflective shaders, that spread the reflection. This slowed down all the rendering dramatically. The shadow samples, shader samples and overall anti aliasing were also set to high so that all the details would stand out more clearly in the final image.
Of course, I used a separate machine for rendering. So while it was rendering the whole image without hair, I was finishing my hair and preparing them for rendering next. The part with hair was rendered later and it took even longer then the whole image without them.
In the end I composed mountains in the background with the whole image, did some colour correction, some levelling and applied depth of field with chromatic aberration as the final touch (Fig.18).
I enjoyed making this image mostly because the creation process was a real learning curve. I have never experimented so much before with any of my images and this is what I'm most pleased with. Of course the final rendering was more like a challenge rather then something really educational and if I had to render it again I would go searching for some alternatives for sure. I'd probably try to "bake" some of materials; do some sort of faking to match the desired result or go deeper by exploring parameters to find a better compromise between quality and speed. In any case, as I tried the hard way first, hopefully this will be something I can do more easily in the future.