How To Create A Stylized Torii Gates Scene With Unreal Engine

Introduction

Hi! I’m Vladislav Svilpe, a 3D Artist with nine years of experience. Unfortunately, I can only name a single project that people would know — Escape From Tarkov. I was a Senior Hard-Surface Artist working on weapons.

That explains how I got all my skills. When you work in no-name startups, you have to grasp every aspect of work that your team is doing, whether it’s VR, indie, stylized, or realistic. I am self-taught through and through, which also comes with the disadvantage of not knowing AAA pipelines and working with my instinct and tutorials on how the big leagues do things. But hopefully, this year I can land a job in those studios (in this economy, funny, I know). So if you’re reading this — hire me, please!

But you all aren’t here because you’re interested in my stories; let’s just jump into the breakdown!

About The Torii Gates Project

My projects tend to be inspired by a single great artwork. Whether it be a concept, a quick sketch, or just an art piece — if I gravitate towards what I see, I can just start doing it. It also helps that I’m a big fan of the colors blue and purple. 

“Torii Gates,” in particular, was a piece that taught me how to learn and practice a stylized environment in UE5, how to make beautiful grass (since it’s the main focus of the entire project), and how to sculpt and texture stylized wood.

For this project, I was heavily reliant on the art itself and had no extensive references except for the wood:

What helps, but I can’t really mention them as references — rather as mood-seekers and inspiration probes — is that I have a collection of some great art with no unifying theme: just pieces I liked, which I can glance through and pick up some details and ideas.

Composition

When working with a concept that has only one angle, the composition is pretty straightforward, especially when it has humanoids — match them first and build off of them. The few caveats are that you have to eyeball the camera’s FOV and perspective, and elevation for the whole terrain, but this comes with iterations anyway.

We can get the scale of the gates, the distance between them, the hills, etc., solely by comparing them to human-sized boxes:

Modeling

For modeling, I use 3ds Max, and a couple of scripts that I can’t imagine working without are:

1. Quad Chamfer
   Simply put, when you’re done with your mid-poly and have every smoothing group sorted, you can quickly make a high-poly version using this script that adds support loops on the edges of smoothing groups. It saves a lot of time.

  2.  TexTools

     Besides some useful UV tools, it has a function that splits your mesh onto UV islands based on smoothing groups as well.

The gates were initially what made me do this scene, as I really wanted to practice my stylized wood sculpting, and when I found the “goat” of a tutorial on the subject, it was a sealed deal.

In the blocking stage, I gave them just a little bit of slants to make things more stylized. After this, it was just a matter of putting it in ZBrush, sculpting, and decimating to make a draft low-poly version that I’ve cleaned in 3ds Max to leave only the topology that affects the silhouette. I also saved myself time by sculpting only front faces and single parts that I could rotate around.

A quick breakdown on rope, also known as Shimenawa — nothing fancy, just a couple of modeling tricks:

The vegetation, on the other hand, was more intricate. While the models themselves are regular, as you would do foliage for games with precise card cuts to reduce overdraws, additional LODs, and mipmapped textures, the setup and master material in Unreal Engine is more interesting to look at, which we discuss later.

Texturing

Since the gates are so far away and details aren’t as visible, I used a quick Stylized Texture Generator with additional masks and dirt overlaid a bit. Also, given that we only see them at one angle, I shaded them directly in Substance 3D Painter because I didn’t like how they were looking in UE with a clear base color. The result can look off in the editor, but it works in context, and that’s all that matters. Alongside, I made a shade mask that I could use to further tweak the look on the fly in the engine.

Now that we’re stepping into the UE territory, I have to state — I’m not a Technical Artist. All techniques mentioned are unoptimized and made solely to look as good as possible. This isn’t a production approach, but may contain implementable ideas. So, take it with a whole Morton’s package — it needs way more than a pinch of salt.

All the foliage uses Material Parameter Collection, with an easy setup I can control World Position Offset and color blend of the wind. Speaking of which, the grass wouldn’t look as lively without the best tutorial on grass from maestro Ghislain Girardot. I mainly focused on implementing WPO from this video.

WPO setup, you can follow the tutorial and it will look the same.

The base for the stylized grass is almost always the same — masked, subsurface, two-sided, specular at 0, fully rough, turning off tangent space normal, and getting Vector3 (0,0,1) as normal.

Base color variation is more interesting. Here’s a list of what I’ve implemented into the master material:

1. Blend/Switch from Runtime Virtual Texture to Uniform Color.
2. Color blend based on the scale of each instanced mesh, subtly making small grass a bit darker.
3. Blend to apply previous settings only to the tips of the grass or override the texture of the grass completely.
4. Hue control.
5. Per-instance hue variation.
6. Wind swipes blend.
7. To shade flowers better, there’s a side gradient — I can control the rotation and strength of the color that is being added. An important addition to the chain is Vertex Interpolator; without it, the gradient will be applied to the whole foliage actor and not per instance of grass.
8. Control to darken the grass that’s far away.

I have five layers of grass being sampled on the landscape for full control — scale, color, density, rotation, and culling for each patch/layer. Only base grass uses RVT to get its color from the landscape; the rest utilize their own shaders with different settings. That’s why some clumps are so sharp and visible while others blend more with each other.

Assemble The Final Scene

I think the most noticeable change to the reference is the overall tone; mine works a bit more bluish, and the flowers do not exactly match the original work. Besides that, I tried my best to make it 1:1.

It’s all about the perspective. When working with one angle, you are allowed to do unorthodox things — anything that’d contribute to the scene’s visuals.

The birds’ models and animation come from 3ds Max and use the Alembic file format. The reason is that it was the easiest and fastest solution — it saves animation per vertex, and you can just import the file and put it onto the UE timeline. I made one bird, animated the wings, duplicated it into a flock position, duplicated it to get the second flock, and gave them a bit of swing back and forth so they won’t be flying like robots in one direction.

For the sparkle particles, I used the Niagara system, which has two emitters. The first one is located as a box above the ground and shoots invisible particles onto landscapes; on collision, they die and spawn second particles with a shader that has opacity multiplied by the wind mask. Thus, they’re only visible when there’s wind on the grass, which gives grass a swiping shimmer rather than just a constant glint like snow. This is probably the least optimized thing in the entire scene; there are constantly hundreds of particles that are not even visible at all times. Although at the moment, it was the quickest solution to spawn sparkles on the grass, I might look into that if I ever use it again in another project.

Lighting

I utilized a classic approach for lighting (key, fill, rim) as much as I could. You cannot go wrong with this approach, and Lumen helped with additional GI and tied the scene together.

From the beginning of the project, I tried to match the reference with shaders and lighting, so there was no particular post-processing added to the sequence. I changed temperature, desaturated slightly, lifted, and tinted the shadows.

Summary

By coincidence, when I checked the dates, I realized I finished it in exactly one month of full-time work. Arguably, I could’ve done it a bit faster, but I had to either remember or learn aspects of UE. If I were to do such a project again, it would be much easier and therefore more fun.

I think all I have to say about the how/why/what is to never give up. There were so many times when I was looking at the results at the end of the day and thought to myself, “Damn, I’ll never finish it; it’ll never be as good as the reference.” But bit by bit, I’ve managed to make it a final piece. Small steps beat no steps. If you’re able to do or learn something, even the smallest, in a day, it’ll make a big difference in a week/month/year.

Thanks to 80 Level for this opportunity and to readers for their time and interest in my work!

Vladislav Svilpe, 3D Artist

Interview conducted by Amber Rutherford