hinterland - using offline renderers to build a futuristic city

Stan Verbitsky talked about the way he uses offline renderers such as Octane and Redshift to build his iconic architectural pieces and the benefits that they provide for his projects.

Introduction

My name is Stan Verbitsky, and I am a 3D artist living in Toronto, Ontario. I've been into 3D since a very young age, although I abandoned it for a while and didn't pick it up seriously again until my 20s. My inspiration for getting into 3D modeling were the video games I played as a kid. I recall playing MechCommander way back when I was 13. Something about the level of detail and fluidity (for the time) of the animations of the 'mech sprites used in the game really enraptured me. Much later, I gravitated toward hard surface environment art – first for games and VR, and more recently – renders. While I am still passionate about working in VR and games (and do that for a living), I also love the speed and flexibility offered by offline rendering. Of course the trade-off is that the final result is static and non-interactive, but not having to worry about optimization or UV's means more time I can put into the atmosphere and artistic direction. 

General Workflow

Another major reason I like working with rendering for my personal art, is that since topology and triangle counts aren't as much of an issue as they are with game art, I can take a lot of shortcuts with how I build my geometry. I created a large volume of the assets for “hinterland” using traditional sub-d modeling, but there'd be times where I'd want more complex compound forms, and didn't feel like dealing with 'proper' modeling, so I'd boolean-combine some simple geometry, then take it into ZBrush and fuse it together using dynamesh with a high blur setting, giving the illusion of a complex hard surface model. I would then decimate the result and export back into my scene.

I also utilized 3DS Max's new chamfer modifier a lot – it has been improved drastically from previous versions, and it is way faster (and more importantly, nondestructive) than placing control loops, especially in cases involving hard surface modeling.

Inspiration and Reference for Hinterland

The mood and color scheme of this piece was actually something I wanted to try out for a long time – ever since I discovered the photography of Marilyn Mugot, in particular, her “Night Project” series (https://www.marilynmugot.com/night-project ).
I love the juxtaposition of, I guess you'd call them “pastel(?)” colors with the rigid and busy forms of urban architecture in her photography. I wanted to do something with a similar vibe, but with the cyberpunk levels dialed up a bit, just enough to give it a 5 minutes in the future feeling. Reference was huge for me, and I spent a lot of time compiling a detailed mood/ref board. I would strongly encourage any beginner artists to learn how to gather and use references instead of just assuming what certain things should look like. I believe reference is king when it comes to hard surface art, more so than technical ability. It's also a good way to avoid the common pitfall of letting your modeling tools guide the look of your work, since if you're working “reference-first” you are deciding what you want your end result to look like, and then using your tools to solve the problem of how to get there. This piece is quite stylized in terms of color and lighting choices, but the architecture and props are still grounded in reality.

Lighting and Colors

I spent a lot of time experimenting with different color combinations and hierarchies – even though I knew what I wanted the palette to be early on, determining how and where to allocate specific colors was still pretty time-consuming, especially since I consider color to be my weakest skill. I've never really gone out of my way to learn lighting principles either, and I conceptualize lighting from a pretty pragmatic perspective. That is, I just treat it as a tool to establish planar separation, visual interest, and hierarchy. From the start, I had a concrete idea of which areas and surfaces I wanted to illuminate and which to remain in shadow, and I also knew that I wanted a lot of illumination spilling out from the open doorway in the foreground. The rest sort of organically evolved from there. Though that is not to say the lighting was quick or easy – there was still a lot of brute forcing to see what looks good and what doesn't. 

Material Workflow

The material creation was by far the easiest part, and the main reason why I like working with offline rendering. This was the first project on which I used Redshift render, and I am very impressed by the power and simplicity of its procedural texturing system. While obviously not as sophisticated as a stand-alone application like Substance Designer, being able to create localized dirt and edge wear without worrying about UVs is a huge time-saver, especially for a scene with as many objects as this one. Another advantage of building shaders in this way is that since you are making use of the curvature and AO of the geometry, you get 'smart material' functionality for free, allowing for tons of flexibility.

This method isn't perfect for everything, so of course you have to pick your battles! One major drawback to this workflow is the inability to place specific details or surface features such as leaks or decals/labels. In these cases, it's usually a good idea to map those details to the object's UV's, and combine with a triplanar shader using material masking. Similarly, if you want to have wider or narrower scratches in specific spots (e.g. more on corners), you are pretty much out of luck. Of course every technique has its drawbacks, but since my work tends to focus more on mood and atmosphere than photo-realistic textures, I find this method suits me more often than not.

I should note that Octane also has a material system that allows for some procedural texturing, as do most renderers. However, I do somewhat prefer the way Redshift allows for a “radius” input for its AO and Curvature. In Octane you are pretty much stuck multiplying the AO/Curvature channel with another layer, giving it more of a 'masked' appearance, rather than the non-uniform blur effect you get with Redshift. I haven't used Octane in a while however, and they may have updated this functionality. 

In my experience Redshift has numerous other advantages over Octane, like being much faster and way smoother with computing advanced effects like volume scattering. However, that is not to say I think one is better than the other. One area where Octane dramatically outperforms Redshift is in its displacement system; while Redshift uses the more traditional subdivision method for displacement, which can very quickly tank live viewer performance and render times, Octane is able to render incredibly detailed displacement at almost no overhead, to the point where there isn't even a need to supplement it with a normal map.

I love this feature of Octane, though even that has its drawbacks and advantages – it seems to be UV dependent, and UV islands that are not connected to each other will appear to 'float'/break apart (I speculate this is because Octane's algorithm uses some combination of parallax occlusion, and displacement geometry only around the silhouette when viewed from a grazing angle). This will not be an issue when using traditional subdivision style displacement, since it is simply pushing out the object's "physical" geometry. As with anything, use what suits your needs and what you're most comfortable working with, but try out many things!