Developing the Scene
As I said, I did not conceptualise the scene at first, and as such I spent time kicking around various ideas for this part. I made a row boat at one point. I think I also made a dock. My big issue with this was getting a crab to appear engaging. It still doesn't, in the end, but at least it's less static then it would have been. My final choice being to set him in a net somehow, I studied Pascal Blanche's work carefully and found that the key was to use shadows to create motion and a sense that something is happening. I repeated that here and feel that it helped greatly. Fig.08 and Fig.09 show several of the ideas I considered for my scene.
Rendering the scene was murder at first because I had some tricks to learn. I'd have a good deal of render power for one scene and it was nonetheless not rendering; it would die or just go on forever. I was baffled for a bit, but then I had an epiphany about two things: displacement map implementation and light map size.
I was rendering at 6144x4608. That led me to look at the edge-length in my displacement settings where I had been using a value of 0.5. Too much edge length combined with too many mesh subdivisions was way over done. I found that, because of the large size, a value of 2.0 was fine. I also took all my Turbosmooth iterations down to a value of 1.
Next there was that light map. Looking at the Mental Ray message window told me I was spending a lot of time regenerating light maps and, sure enough, I had left it set at 50% of the render size in all my materials, so I calculated that a light map at 50% of a render at 2048x1536 would make a 1K light map. My renders took so long because I was making nearly 4K light maps for every material. I cut the value down to 14.8%, which made roughly 2K light maps. With these improvements, renders came down to about two hours.
Post Production: Photoshop Techniques
I brought it all together in Photoshop. I was rendering to OpenEXR. This let me build up layers of renders with different exposures in Photoshop, where I selectively masked to get better values and colour. Another trick were the shadows. In addition to the OpenEXR, I had rendered out passes, but the shadow pass was a bit choppy for some reason. My solution was to take an earlier greyscale render and use it as an inverted mask in Photoshop to effectively create a shadow layer. This proved a powerful technique and let me adjust the shadows very freely as a result. I did something similar to tweak the specularity (Fig.10).
This is one of those works that I could have (and should have) spent much more time on. I would like to have seen what I could do to really bring this creature to life in a more appropriate environment. I might mess with that someday. But if I could point out the most important thing I learned on this, it is to spend more time in the planning phase - doing thumbnails and concept art and working with colour combinations so I have a better sense of where the project is going. If you know where you're going, you can try to get there more directly.
Aside from that, I feel like I'm at a stage right now where I can tackle projects of this scale more readily because of what this one has taught me. I'm much more enthusiastic about Mental Ray now, and even 3ds Max, but it took a long time to get to this point. Sometimes it seems like 3ds Max gives you all the plans and many of the parts for a great 3D app. and you just have to figure out how to put them together into what works for you. But I'm there now. A word to the wise: to work better in 3ds Max, make a custom Quad and make Maxscript a cornerstone of your toolset.