If love it, make it free (c) ;)
Trying to steal Vray's thunder.
Sylvain Picard showed how he builds amazing 3d environments with scanned content and SpeedTree vegetation.
My name is Sylvain Picard and I am currently finishing my bachelor degree in 3D Animation and Numeric Design at NAD University in Montreal. I have worked on the Contra reboot called Contra 2028, a wave-based TPS done in 8 weeks with 7 other students. My latest project is Tobhtaichean, a contemplative natural environment with some ruins and a fortress. I wish to be and Environment Artist / Level Artist.
The main idea behind Tobhtaichean was to create an environment that would bring people to fantasize and allow them to contemplate the nature intertwined with manmade elements added from various periods. Another goal I wanted to reach for this project was to improve my skills in modeling, texturing and shading by using various workflows and by developing skills in different software solutions.
Being a one-man project, the scope of the environment itself was a big challenge.
I have to reassert the breadth twice during the project. The first major project change was to go from a render-only project to a playable one. The next step was to avoid any modern detail and put more time creating the landscape and designing ancient architectural elements.
What I liked about this project is how much I learned and the new skills I gained in the process. Creating the foliage, sculpting various assets and working with SpeedTree are a few highlights I can think of.
The landscape creation was probably the longest thing to do because the entire project was going to be made around it. I used mainly the software World Machine to create the landscape. I had about five different versions of the landscape from the same design. One design detail I wanted is to put a big lake between mountains to build a nice vista. I always have to keep in mind to keep the shape and proportions in check to ensure that we have the impression the world I create feels big.
When working with such a big environment, the proportions and size are critical. If you do not plan correctly, your various assets may look out of place. Another problematic area was with the actual mountains shape: I was not able to have them look right. To fix the problem, I had to make the shape first and transfer them as assets in Unreal Engine 4. This method allowed me to create my landscape and to have the freedom of layering out mountains where I wanted them and how big they were. Unreal has a system for the landscape to optimize geometry based on camera distance. Hence, it was easy to avoid high polygons count from the landscape itself. For the mountains, I used the Unreal LOD system that reduces the triangles count with the distance.
SpeedTree was a great software to work with. It is user-friendly, has many options and even if it mainly procedural, you can manually tweak any part of what you create! The first step was to create the meshes for the leaves in Maya and import them into SpeedTree. This was to have a better model than the one generated by default in the software. The bark textures for all three models were made using Substance Designer. I did two variations of each tree species I wanted, so it there is six trees total; two maples, two birches and two pines. Inside Unreal, I have added further shader variation for the leaves resulting in hue and base color variations.
Qixel Megascans assets were used for the leaves, flowers, bushes and grasses textures. Sadly, here in Montreal, there are no leaves on the trees and the grass is mostly covered in snow for six months, so I had to find another way to get foliage textures for my project. I wanted to restrain the use of Megascans because I did not want to rely too much on pre-built assets. I created most of my assets and textures with Maya, ZBrush and Substance Designer and Painter.
Photogrammetry techniques could have been a nice way to proceed, but it would have probably clashed with my assets. Photogrammetry and “handmade” assets do not seem to come from the same universe. As previously told I used Megascans for the textures, but I have to tone them down to better fit with my environment.
Working on such larger scale project, you see many things that can or could be done but you have also to think in term of optimization. As I said earlier, I never thought of making it as a playable environment, but as my project was evolving, I saw the potential to have a more interactive project for people to explore. So one thing that helped with the fps was creating LODs for the trees. Fortunately, SpeedTree has a system that helps you with that. I went from 15 fps to between 30 and 40 fps at Cinematic settings. Creating a master shader for the trees helped mainly with the draw calls but have almost no effect on the frame rate.
The lighting was a work in progress throughout the project. I did not need much from the lighting, just a Directional Light, a Skylight, Post-Process Volume and Volumetric Fog. What I really needed to make great was the fog and almost one day out of three I worked on the fog. I really wanted those cool fog beams while having some kind of light propagation. I had both, but they were not working too well together, so I worked without the LPV and got the result I wanted. I added a light function that mimics the clouds passing in front of the sun, so it darkens certain regions of the map. A little detail that adds to the feel.
It took me from three to four months to create that project from scratch. It is always hard to start a project, but it made a bit easier by creating a schedule for my deadlines. Working alone was obviously a big part of the challenge. When you work with a team, it is easier to compensate when some area of the work is behind or not as strong. However, there is also a positive aspect to working alone: you do not have to compromise to someone else idea. I still relied on critiques from my friends, family, teachers, and professionals, as it is important to have different points of view. Having a fresh perspective is invaluable and helps to put your work on the right path. However, the biggest challenge remains the landscape creation, from that point everything went smoothly.