Creating 3D Fan Art Of Elite Imperial Soldier From Warhammer 40,000

Introduction

Hello readers, my name is Hakan Kamar, and I'm a 3D Character Artist. It's my pleasure to write about my latest finished project: fan art of a Tempestus Scion from the Warhammer 40,000 universe, owned by Games Workshop.

A bit about me: I started this journey in 2013, when I enrolled at the Grafisch Lyceum in Utrecht as a Game Artist. At the time, I didn't know what I wanted to do after secondary school, and what better course for a 16-year-old avid gamer than Game Art? I've always been an outside-the-box thinker, which helped me a lot, even if I didn't realize it at the time.

During school, I wasn't 100% sure what I wanted to specialize in. For a while, I was more drawn to concept art than 3D, hence my internship as a Concept Artist at Guerrilla Games in 2016. Looking back, that period instilled a design mentality I still use in my day-to-day work, from dissecting existing designs to solving design problems when translating a concept into 3D. After graduating, I decided to pursue a career in 3D art, specializing in characters. Since then, I've contributed to Horizon Zero Dawn, Arctic Awakening, and, most recently, as Senior 3D Character Artist at PlayerUnknown Productions, Prologue: Go Wayback!

Over the years, I've noticed that some skills develop in a linear way and others don't. Learning a program is linear: do it X times for X days and you'll progress steadily. Results may vary, of course, but that's how it works in my head. Learning to sculpt is a different story. There's only so much to learn in a program, but skills like sculpting, drawing, modeling, and texturing involve a bit of "magic". Developing that magic is what I consider most important: being able to see why a design, a texture, a piece of topology, or a model works, or when it's good enough to move on.

Programs complement those skills. You learn ZBrush because it makes sculpting easier, you use Substance 3D Painter because it makes texturing easier, and so on. The saw and hammer merely enable the carpenter. I learned mostly with paid tutorials, like Rafael Grassetti's, free tutorials on YouTube, anatomy books, observing the real world, and the modding projects I've joined over the years, all of which have proven very effective. The hardest part, I'd say, is the sense of isolation when you're starting out. That's why I joined modding projects like Skyblivion and Rise of Mordor.

The Tempestus Scion Project

When you're starting out, you try making 3D models and characters but feel like you're missing the understanding to even critique your own work effectively. You can do random projects and hope some of them unlock the part of your brain that makes you think "the knees need to be lowered" instead of "why does this look weird?" Or you can take a more precise approach and ask yourself: "Which skill feels underdeveloped in my toolkit?" Then pick your next project to push you on exactly that.

With that mindset, I've set out to learn or focus on a specific aspect of character art with every project, be it anatomy, armor, or clothing. Approaching it this way makes the whole learning process more manageable. With that in mind, I decided to make the Tempestus Scion as an exercise in a character with more hard-surface elements. Anyway, enough about me. Let's start dissecting the piece.

The idea of doing a Tempestus Scion started when I got fascinated by the 40K universe a few years back. At the time, I didn't have room for something like this; I was already working on one personal piece and wanted to finish it before starting anything new. That moment came last March. I was looking around for something to make in that universe, and a friend suggested a Scion.

I looked online and found the image below. It's a piece that just clicked with me. I thought, yes, that's it. It also fit the fact that I had no characters in my portfolio that leaned heavily on hard-surface elements, so this one was perfect.

References

It's an interesting concept to tackle, because this character is part of a larger IP, so I had to understand the lore behind it as well. That research adds depth to your understanding of the character, and it adds more checks and targets to hit. As an example, a quick look at the 40K wiki gives you this excerpt:

Good! Now I know I can look for Hot-shot Lasguns, Hellpistols, Carapace Armor, and Omnishield Helms. This helps a lot. To explain it a bit further: a concept that exists on its own, without supporting lore, sits on a single layer. You see a trooper from a universe that doesn't exist, so the concept art becomes the most important source of information on what you need to make. The why of everything becomes less relevant.

Once that concept art exists in a world with lore you can access, it adds a second layer. All of a sudden, everything the character is, has, and is doing becomes meaningful in a different way. The weapons they carry have their own history and use, supported by the wider lore rather than by a single image. This also allows you to morph or change the concept a bit.

To tie that back to my piece: these Scions aren't wielding Hellpistols, since I don't see any in the original concept, and I can see there's some kind of planetary drop going on. It made more sense to me to add a larger, machine-gun-like weapon alongside the Hot-shot Lasguns.

Most of the reference available for a Scion comes from miniatures, and miniatures are exaggerated in shape and form by design. That reference is helpful for working out what goes where, reading the larger shapes and picking out the main colors, but it loses its effectiveness from a texturing point of view. Since this piece is more realistic, a one-to-one copy would have made it look toyish, and the scale would have felt off. This is where that "magic" comes in.

Going a bit deeper: before I start anything, I always collect references. Not too much, though. I find it difficult to understand what I'm chasing if I have too much of it, so I keep the amount low, and I'm very picky about which images make the cut. Too many images can muddy the waters.

If I have ten different rust references, I might end up mixing them all together. I might not realize I'm blending sea salt erosion with rainwater erosion on different types of iron, and now my rust isn't either one of them.

With reference in place, I like to start with a strong base mesh that already sits close to the proper proportions. I'll begin with a blockout and build from there.

Usually my intention isn't to copy the reference one-to-one, but to base my decisions on it. Sometimes it's also about validating a feeling I have when I look at my character: do I get the same feeling as when I look at the reference? If so, I know I'm approaching that same "magic".

While collecting references, I'm also thinking about my deliverables for the project; in other words, what I want to end up with. I knew I wanted to recreate the scene from the main concept I was following. I also knew that, even though the character wears a helmet, I wanted a proper face underneath that you could see with the helmet and balaclava off. And I wanted two types of guns and two types of backpacks. The list became:

• 1 full armor set;
• 2 faces;
• 2 backpacks;
• 2 guns

Modeling

Now that my reference has been collected, I can start on the blockout. My main goal with the blockout is getting the proportions right and working out where everything sits: whether the shapes and sizes make sense. I'll keep making larger changes at this stage until I stop seeing major errors. I do all of this in Blender, without touching any other program.

I like to start from the 3D Scan Store male base mesh. I'd say it's important to understand when you want to build your own "human" body and when a base mesh will do. For this project, I wanted to hone my hard-surface skills, so I opted for a strong base mesh.

A note from future me: matching the initial character to the UE5 mannequin is good practice, since it lets you drop in animations later without too many workarounds.

My goal in this phase is to "force" the proportions. Even though the anatomy underneath the armor and clothing doesn't fully make sense, I still push it towards what I want the character to look like in the end. It's the blockout. You're the artist and king of your own castle, so don't worry that it isn't there yet.

I don't aim for 100% correct on everything at this stage. Getting it to a point where I can "see" the character is good enough for me. Getting the proportions right is what I go for first. Once the blockout looks right from afar, I zoom in, refine the smaller elements, and see if I've missed anything. This is where I just go with the flow, adding and removing elements as I go. At this stage, it's still mostly box modeling inside Blender.

Between stages 2 and 3, I did a Marvelous Designer pass on the clothing underneath. To get the clothing to react and fit snugly, you can morph the avatar inside MD onto your updated character. Basically, you import the avatar again, but as a morph target, and the cloth follows along. Here's a good video explaining it:

As I refine the blockout, I'm constantly fighting the miniatures and the concepts. Some shapes in the concept are hard to reproduce, since it's not really a character sheet but rather an illustration of the character in its environment. That tends to exaggerate or shrink certain elements in unnatural ways. Here is the same progress for the backpack:

A short bit about my Blender add-ons. I mainly use Blender for hard-surface modeling and as the main program where all my meshes come back together. Marvelous Designer cloth done? I check it on the character in Blender. High poly in ZBrush done? I check a decimated version in Blender. The program's versatility and its ability to handle lighting without performance problems are king.

I have some custom scripts I use for fast exports and fast imports. I also quickly make materials. Once I was happy with the blockout, I could start refining it in ZBrush. ZBrush is where the "sauce" gets applied. I love ZBrush, and I feel I get closer to expressing myself there than in any other program. Everything I did before this was laying out my canvas: planning and anticipating. Now, in ZBrush, it all comes together.

So which add-ons do I use?

• Hardops 26: Lead
• Boxcutter 26: Stiletto
• Zen UV V5.4.0

My goal in ZBrush is to make the materials read without textures: hard metal should read like hard metal, and soft metal like soft metal. You can see that the harder blue metal has a different finish than the softer gold metal. I imagined Carapace Armor to be made from a composite rather than a single material.

My main brushes are hPolish, Dam Standard, and Standard. Nothing special, really. For metal, I'll initially go around the edges, wear those down, and then work towards the middle. Depending on your metal, the amount of work needed in the high-poly can be quite minimal.

The blue metal is something I eventually changed; I removed all the larger damage. As I mentioned earlier, when adding surface detail like this to a real-time model, you need to decide whether you'd rather add the damage in ZBrush or through height maps in Substance 3D Painter. For the most part, it's personal preference, and it depends a lot on what the damage is. I still like adding some in the high poly because it's fun, but mostly I leave it for Substance 3D Painter, since it's much more versatile as a height map.

Sometimes I like to take my work mid-process and render it in a different program. This one was done in Blender. Seeing your meshes in a different program and lighting setup helps you better see the changes you need to make, and gives you a better feel for how finished the mesh really is in its high-poly phase.

Once I was happy with where the armor was heading, I got started on the heads. Because I wanted to make two backpacks and two weapons, my plan was to make two heads: a private of some sort, someone with less experience, and a sergeant, the one in charge of the squad and more of an older type.

I started off with 2 base heads from Texturing XYZ. The main reason I use these is to eventually get their skin detail onto the character. I use them as a starting point and deform and shape them to my liking. The most important part is preserving the UVs, so I don't change any of the topology and only adjust the larger shapes at first.

The largest change I made was adding some symmetry back in, so the sculpting would be easier going forward. These heads ended up being quite realistic. After seeing that, I wanted to stylize them and push them towards more of a Darktide vibe. After some feedback, I managed to stylize them further by making them more memorable and exaggerating some of their shapes: bigger ears, eyes, and lips, stronger cheekbones with more pronounced zygomatic arches, and some deeper skin folds.

And while I was doing that, I also made sure to sculpt more on the backpack assets.

The blankets attached to the tops of both backpacks are identical and get reused on both assets. The only thing I needed to figure out was how each one attaches to the backpack's structure.

For the high-poly phase, I think it's important to understand what you're doing and what you want out of the high poly. If you really break it down, you could say we're painting data. All the shapes, details, and finishes we create are great data sets for the eventual normal maps, curvature maps, ambient occlusion, and any other texture maps you want to bake. Because that's what they become in the end: texture maps. Texture maps that present themselves in pixels. Pixel density is something you need to be aware of. Adding more detail to your high-poly model might not work out in the end, as there are simply not enough pixels per dimension on your model to accurately represent what you are sculpting in.

So when you're going all in on your sculpt, ask yourself: do you really need to zoom in this far? Do you need to work at this high a subdivision, making scratches that won't even register in the bake? Could you delegate some of those details to Substance 3D Painter? Realizing this helps you find ways to speed up your workflow, come up with your own tools and, maybe most important of all, know when to stop working on the high poly and move on. For me, that point was when I added surface wear to the body. I didn't need to do that in the sculpt. It could have been done in Substance 3D Painter, since the height maps I'd use are exactly the same.

Then, for some, the fun stops. For me, retopology can actually be therapeutic, as long as it doesn't take a long, long time. With the high poly finished, I take a good look at it and decide which parts I want as separate pieces and which I want as one mesh with its own flowing topology. Exploding the model into pieces like this makes my life a lot easier. If I tried to retopologise it all in one go, I'd be making it unnecessarily complicated for myself.

Retopology

For this project, I wanted to see whether retopology in Blender could compete with Maya. I used to use TopoGun as well, but found it didn't do enough to justify being a separate 3D package. In Blender, I found an add-on called RetopoFlow. It has most of the tools Maya has, and for this project it served its purpose. Another thing I noticed is that Blender is a more stable program than Maya; I remember Maya crashing far more often. I'd still say retopo in Maya feels better to me, but the fact that Blender doesn't crash and has better lighting makes it a hard choice. If you haven't tried Blender with RetopoFlow yet, give it a shot and let me know what you think.

I don't use a lot of tips or tricks here. For me, retopology still comes down to fairly standard tools. What you're retopologising is usually so specific and precise that any trick falls short, and there's always some cleanup left at the end. It's very easy to overlook a double vertex or a bad normal, or to leave a quad untriangulated, only for the engine to triangulate it the wrong way. I'd rather keep the reins in my own hands and do the retopology myself. It's one of the most tedious parts of character modeling. During the blockout and high poly, it feels like you're cycling down an awesome hill. The moment you hit retopo, you have to slowly walk back up that hill with your bike.

With the retopo done, I could move on to UVing all that hard-earned topology. So now I have the high poly and the low poly models, and I can dedicate time to unwrapping. For me, this stage isn't only about unwrapping the model; it's also about making sure the mesh I built is sound: no weird faces, no non-manifold edges, and ready to bake correctly.

I tried unwrapping in Blender, RizomUV, and Maya. First, native Blender. Coming from Maya, I found the native tools extremely lacking. I missed so many actions and tools that I gave up on unwrapping this project in Blender almost immediately. What I missed most was Maya's Unfold along U and V, which makes it super easy to get straight UV shells.

I also wanted to try RizomUV for this project, since I'd heard a lot of good things about it. I won't go into too much detail, but I'll say this: Blender is a straight-up no-go for me. Maya is great for anything from simple to fairly complex, and covers most of what Rizom does, but Rizom gives you a bit more control, since it's a program built specifically for unwrapping. If you have lots of texture sets and UDIMs, I'd say go for Rizom; at that point it's safe to assume the project is complex enough to warrant it. If you're not at that stage, go for Maya. Its UV tools are simple but very effective.

Texturing

And now we get to bake our details onto the low poly mesh. Baking is super satisfying for me. It's where the art and the technical side meet properly for the first time in the process. They've met before, during retopology, but now they really interact with each other.

I do all my baking in Marmoset Toolbag. I've been using it for years now, and its intuitive workflow is a real boon. Before that, I used xNormal, and I never liked how baking was a black box there. With Marmoset, I can actually see what's going on and make quick changes. I keep a separate Marmoset file for each texture set. On this character, I had three for the main armor, a fourth for the sergeant, and a fifth for the private. They all have a similar setup:

Each separated asset from before gets its own bake group. That way, whenever I need to make a change to the topology or UVs, I just re-export the mesh, and it refreshes automatically. I hit bake and, ta-da, it's in. I think this kind of non-destructive, organized workflow is so important for keeping the creative mind intact. You don't want to sacrifice iteration because the process is too tedious. The most important maps I bake are:

• Tangent-space normal map (duh)
• Curvature
• Ambient occlusion (I bake two, at different ray distances, so I have a close and a far AO to choose from in Substance Painter: the close one for small, tight dust build-up, the far one for broader, all-encompassing dust)
• World-space normal map
• Vertex colours/ID map
• Convex and concave maps

I check for artefacts in Marmoset, but I also set up the character in Blender to see it in different lighting and in a different program.

This is where I inject a lot of life into the character. I start texturing once I'm happy with my bake results. At that point, I make a specific "texture mesh": a mesh laid out so I can easily see everything I need to texture. Here you can see the guns on one side, the cable connecting the gun to the backpack, and the second backpack off to the right. Basically, it lets me texture everything in one file instead of several.

I always try to get all the major color blocks in first, then slowly work through each of them, adding rust, metal wear, and any other changes I want. I usually don't stay on one material until I think it's done. Since the materials play off each other, I think it's important to hop between them often, so you don't overwork any single one, and everything stays cohesive.

Here I'm using the gradient filter to create an anodized metal look. It works the same way as the Gradient Map in Adobe Photoshop. I've used it for the anodized look, and also to add subtle paint discoloration to the edges of my armor, as shown below. I can either paint the map myself or use a thickness map baked in Marmoset Toolbag.

Here I'm using the same technique, but instead of painting or placing a gradient map, I'm using a baked thickness map to add subtle color variation closer to the edges of the armor.

Here you can see how I "paint" the paint wear near some of the armor pieces. The mask stack is quite simple: I start with the curvature map, blend it with a generic noise, and add a paint layer on top. I finish off with a Levels adjustment to make it crisp. This lets me paint on the paint layer at any opacity. The Levels keeps the result crisp, while the noise underneath adds subtle variation to anything painted at lower opacity.

The armor was one part, but the faces also needed some touch-ups. Because the heads started from VFace scans from Texturing XYZ, I already had a super solid base. My goal in texturing was to make them a bit more stylized by exaggerating the colors around the eyes and lips. I added some reds, blues and yellows, touched up the lips, and made the normal detail a bit crisper by using the curvature as a height map. And that's about it. Because VFace gives you such a strong starting point, I don't need to touch it that much.

Grooming

I wanted to dabble a bit in hair production. The characters are military, so my aim was to give them short buzz cuts. I didn't want to go through a full hair-card pipeline for the head hair, but I was willing to do that for the eyebrows and eyelashes.

The hair deliverables came in two parts. The first was the hair cap: a textured mesh that sits between the hair cards and the head mesh. The second was the hair cards for the eyebrows. Since the head hair and mustaches were going to be buzz-cut, I knew they didn't need hair cards, so only the eyebrows did.

I make hair caps in Maya using XGen. Once I've gathered reference that gives me something to aim for, I start by separating the part of the head mesh that isn't going to have hair.

The result at this stage isn't something we can use yet. Even though the hair is there, the normals aren't facing the right way; they're facing the camera. To fix this, we need to convert the XGen description to an Interactive Groom. Once it's a groom, we can align the hair normals with the surface of the mesh.

Once that's done, we can convert it to a mesh and bake our maps onto the cut-out part of the head mesh that's visible. I bake basically all the maps from this and do the color in Substance 3D Painter.

I could have spent more time finishing this, but I didn't want to put much more into something I'd be covering with a balaclava. In production, you wouldn't spend too much time on this either, since 90% of the time the character is wearing either a helmet or a balaclava. I'll touch on one final thing: the eyebrows. I won't go into much detail, since the workflow is the same as any standard hair card workflow: bake strands into textures, make card strips, and place them on the mesh.

I used FiberShop for the strands and the Groomer's Tools plug-in for Maya. FiberShop can be swapped out for XGen when creating hair cards, and in the future I don't think I'll use FiberShop, as I ran into a lot of weird bugs and crashes with it. It might even be discontinued. Groomer's Tools, however, is a lifesaver. It adds so many quality-of-life features for making hair cards that the whole process no longer feels daunting.

Posing

Now we have hair out of the way, the character is finalized, and now we only need to get it moving and posing. I pose in Blender using an add-on called Auto-Rig Pro. If you only need still poses, it lets you get a working pose very quickly. I haven't tried integrating it into an animation workflow yet, but for static poses it works very well and saves tons of time. In the past, I could never quite get my characters posed the way I liked, but with ARP it's so much better.

For this project, I had two rigs: one for static poses, which you can see above, and the Unreal Engine 5 rig. I skinned my character to the UE5 rig to see how it would react with physics assets and how well I could get the gun's cable working, since that was something I was concerned about.

I knew I wanted two different guns with two different backpacks. One of the guns has a cable running from the magazine to the backpack, which was something I wanted as well. I eventually got it working by making a custom rig where each bone between the gun and the backpack has its own physics sphere, and with some tweaking of the settings, it worked. I'd love to go into detail, but that would mean untangling how Unreal Engine handles physics and how much I struggled with it at first.

The animation works using the Advanced Locomotion System. Skinning the character to the existing UE skeleton is what makes this possible. The gun took a bit more time to get working, and I used Claude to help me figure that part out, writing custom scripts and vibe-coding some tools to make it all work. For the rig, it helped me a lot in understanding how to bring an assistant like that into my workflow purely as a coding helper: someone to help with implementation and with making custom plugins and scripts.

A short note about being a Character Artist for games. When it comes to game development, I'm always looking beyond the character. My job doesn't start and end with building the character from scratch. I see it as my responsibility to also think about animation and concept art; when I'm involved in what comes before and after, I can be most effective. At the end of the day, the job isn't to create a great character per se. It's to create a great character that fits the world and the development cycle as well as it can.

Here's something I tried: a very simple, generic eye and blink animation made purely with code. It doesn't need to be special, just something running while the character is idle. It works by adding random movements to the eye and eyelid bones, with random intervals in between. Notice the texture and how I was messing about with this while the texture was not done yet. Procrastination!

And from procrastination! We go to the presentation.

Rendering

For the presentation, I knew I wanted to recreate the original concept in Unreal Engine. I also wanted poses with the different backpacks and guns, renders of the heads, guns, and backpacks, technical renders, and, finally, high-poly renders. To plan what I needed, I split it up by program:

• Unreal Engine: main scene, character turntables, gun renders, animation tests;
• Marmoset Toolbag: technical renders;
• Maya (Arnold): high-poly renders

The main scene in Unreal Engine was definitely made with a concept artist's mentality. If you zoom out, the scene makes no sense; everything is placed purely to support the shot. The ground uses a random muddy Quixel texture, with a Quixel tree asset scattered across it to give me something poking through the ground. The smoke assets are basically planes I placed around the level wherever I wanted to control the composition.

The smoke assets are images of smoke slapped on a plane and placed accordingly, same for the explosions, muzzle flashes and the orange streaks going through the air. The streaks are soft circles stretched out using a very simple shader.

That's how I built my scene in Unreal Engine, but I still had to render the turntables. To render anything in Unreal Engine, I had to enable a plug-in called Movie Render Queue. It lets you queue up renders, and it's a very helpful tool if you want to render anything in Unreal.

Let's dive a bit deeper. I have a pedestal in my level with some drapes behind it that I made in Marvelous Designer. The pedestal is set to turn 360 degrees over eight frames, which gives me eight images, each 45 degrees apart. From those, I can easily create simple GIFs. I prefer this over a few seconds of the character spinning continuously. I think the eight-image method is easier on the eyes and gives you time to look at each angle.

So there's my pedestal: anything I place on it turns with it. I have a Level Sequence set up for each camera. In this example, I'll use Sequence_Front, which is the sequence and camera setup for front-view renders.

Selected in the Movie Render Queue window, it looks like this. Notice how the configuration is set to Preset, with my own UltraSettings preset selected.

The most important setting is probably having the Path Tracer enabled. This setup lets me quickly swap out models and poses. Once I have these renders, I set up the technical renders. I don't do these in Unreal Engine, because I know from experience that it takes a workaround to render a flat, unlit model: you'd have to plug each texture into a special unlit material. Marmoset Toolbag is much more intuitive for this, so all my technical renders are done there.

Simply by pressing Add New, you can select which pass you want to render. And finally, there are the high-poly renders. These were inspired by the high-poly renders from the God of War games. Rendering with Arnold's curvature shader (aiCurvature) is a great way to show off the detail and depth of a character sculpt.

Preview of my scene, for the full body renders, it’s identical.

This is what the shader setup looks like. All it does is combine a small-detail and a large-detail curvature pass with an AO node, while clamping the output. Without clamping, the result is pure black and white.

Conclusion

I hope this was an interesting read. I'm always shocked by how much thought goes into making one of these characters. Even though I did my best, I had to leave out so much that this didn't turn into a 40-page essay on the topic. I have a tendency to zoom out of what I'm making and look at it from the broader perspective of character and game art.

This project took far longer than expected due to some unfortunate life events. Personally, I don't like to take a long time on projects; I have so many ideas I want to explore that halfway through any project, I'm already thinking about the next one. I appreciated the feedback I got throughout the project, one piece of which pushed my face towards a more stylized approach. The fact that the industry is burning, along with the uncertainty around AI, complicates everything and adds a sense of urgency to everything you do. At least, that's the case for me. I find it increasingly important to slow down and enjoy the work, not the outcome.

When I first started doing anything art-related, I did it as an escape and because I liked doing it. I think it's undeniable that once you start earning money with art, you find yourself sacrificing some of that joy, because it's become more serious. I strongly believe that part of growing as a professional artist is finding your balance and your own relationship to that shift. There's no straight answer; art is inherently very personal, and it's impossible to give advice that fits everyone.

For me, what worked was focusing on the work. You can zoom out, see yourself starting a cool new project and envision the euphoria of the finished piece, but that can backfire. In a world as chaotic as ours, you don't always have the power or control to finish things. Sometimes you just lose sight of what you wanted to do, and suddenly you're in a creative swamp you feel you have to trudge through, because you want to finish something. Focus on the work. There will be projects you start and never finish, and ones you do finish. No matter how you look at it, all of them involve the work. Focus on the work and the rest will follow. The process matters far more than the final piece. After all, the final piece is the culmination of your process.

Again, thank you for reading!

Hakan Kamar, 3D Character Artist

Interview conducted by Emma Collins