Discussing with Digital Production, the team at Mondlicht Studios shared insights into their experience with Substrate. They covered the various applications, benefits, and challenges of using Unreal’s new material system. hey utilized it to create a two-minute showcase video featuring the BMW M8 Competition, demonstrating the advanced capabilities of Substrate materials in action. A part of the video was presented at 2023’s Game Developers Conference (GDC), highlighting the potential of Substrate and showing off Unreal Engine‘s extensive next step in real-time materials.

by Stefan Evrad (Head of UE Department),
David Schaefer (CEO),
Dmitriy Glazyrin (CEO) and
Zhanna Travkina (Creative Producer)

Always check official docs

This may sound quite boring and very ob­vious but that’s the first thing one has to do (and we know how many are not) to at least to get an in-depth understanding of what Substrate can do and how to handle the new system. The official documentation can be found here. In short terms, Substrate allows you to stack materials on top of each other in an accurate way, and introduces new capabilities within a material. Previously in Unreal, we had the option to use a „clear coat“ material, which contained two layers of material, the top layer mostly designed for varnishes. With Substrate how­ever, you can stack as many layers as you want, multiple coats, multiple bases, and blend them in many different ways. One thing we particularly like is the dual specular which is in each layer, meaning you can have two slightly different roughness levels on one single layer, which helps getting closer to real life materials and offers more creative control over your materials. We used this extensively for the roads and concrete materials, in order to get a nice variation in roughness. We were very happy with how natural it looked considering everything was real-time. In the two images here you can see the really nice specular, which looks much like a crazy calculated rough reflection, but it’s real-time and just looks great.

Finally frosted glass! Or: Rough translucency + tinted glass

Glass has always been a bit of a challenge in Unreal. Correct or at least good looking transparency/translucency was always tricky. This is awesome now, but even much better, we finally have frosted glass. In high demand in so many real-time productions, the impact of this is hard to understate. Changing the roughness of translucent materials now actually blurs the background and also lets you tint them very nicely. We could do this using the thin translucent shader before, but with substrate everything is possible without having to switch shader types. We created these 4 panels that ended up being triggered during the animation to light up and display the BMW M colors. To those who haven’t used Unreal in the past it may seem not as great, but it’s mind
blowing.

Thin film

Things like soap bubbles, camera lenses and for example also some automotive head­light cover plastics on cars have this slight rainbow effect that we call thin film. This was implemented into Sub­strate, and combining it with the tinted glass possibilities. We used it for the headlights to give them a bit more thickness and life, boosting it a tiny bit more than real life to compensate for the fast movements and camera shakes. We wanted to show what’s possible. Usually creatives try to get rid of distracting visual features. Even if they are real, they are most of the time not wanted. We wanted it.

Car paint materials

Perfecting car paints was a crucial task. As subtle as they are, layers of flake, orange peel, specularity, depth, color fall-off caused by the clear coat and clear-coat itself, all should work together to bring complexity and depth within the material. All this was of course much easier to do with the new layer system. Stefan, who had developed his own car paint shader over the years, found it somewhat bittersweet to move away from it. While Substrate made his custom techniques redundant, it also enabled him to design a new shader with far more adjust­able parameters. One main thing that also comes with this system is the fact that it is returning to a specular workflow; meaning that while before Unreal used a „metallic“ value, now it’s the Specular value that will determine the reflection IOR, similar to other non-real-time render engines. Tinting the specular value, once again similar to other render engines, allows you to make materials like gold or metallic paints much nicer. Don’t worry if you’re too used to the traditional method though, there are conversion nodes you can use!

One thing Stefan and any good artist liked to do back in his VRay days was to make sure glossiness values weren’t absolute, meaning 0.95 or 0.05 looked more realistic than 1 or 0 values, as in real life we rarely have absolutely sharp reflections. In Unreal this wasn’t possible before as the roughness value would jump from pure gloss to mat, but now small values work beau­tifully, and while it’s a tiny change we still like knowing that we can very slightly blur reflections just like in render engines.

SSS

Sub-Surface-Scatter materials were already possible, but now have become even better! In order to showcase this, Stefan created a waxy looking version of the car which just looks smooth and very interesting. It was easy to add details and the intensity of the subsurface to create a specific look.

This new system introduces a variety of additional nodes and techniques worth exploring, marking the beginning of expanded capabilities for material creation and layering. Considering all this is real-time,
the instant feedback you get while now having extremely powerful tools really pushes Unreal one step further in terms of visual fidelity, and we’re excited to see where it will go!

Mondlicht Studios is a CGI Studio located in
the heart of Germany. Boasting a significant background in the automotive sector, their ex­pertise is a testament to their deep roots and understanding of the nuances in this industry. The foundation of the company is built on the rich experience of their co-founders, each bringing close to 20 years of professional expertise to the table. 
Surrounded by a small team of veterans and professionals the team is delivering award winning content, providing aid and helping clients to step up. Among these clients are Lucid Motors, LVMH, Nissan, Panerai, BMW, Netflix, Amazon, Netflix and many other renowned brands. 
Over the years, they‘ve been honoured with numerous awards including Promax Global Excellence and Clio Entertainment Award. The team‘s dedication to excellence has also led them to be named among the top 200 digital artists in 2021, 2022, and 2023 in the Lürzer’s Archive.
Mondlicht Studios has also been celebrated multiple times at the Motion Design Awards and in 2023 became certified service partners of Epic Games. Beyond their achievements,
they believe in giving back to the community. Members of the team have had the honor of serving as judges at renowned platforms such as the NY Advertising Awards and Promax
North America. 


mondlicht-studios.de

And our four Authors are:
Zhanna Travkina (Creative Producer), Stevan Evrad (Head of UE Department),
Dmitry Glazyrin (CEO), David Schaefer (CEO),

The post Unlocking Substrate: Insights for Unreal Engine’s Material System first appeared on DIGITAL PRODUCTION and was written by Mondlicht Studios.

by Jonathan Braun (Head of Data), David Schaefer (CEO), Dmitry Glazyrin (CEO) and Zhanna Travkina (Creative Producer)

Creative CGI studio Mondlicht Studios invites you to take a closer look at how to get things from the drawing table to your framebuffer without breaking your pipeline.

Step 1: Getting engineering data

The work starts right after the client contacts you about the project, because the first step is to ask questions or otherwise you might end up burning the midnight oil in misery. It is pretty obvious that you need to ask about the scope and that creating one nice still image is easier than setting up a pixel streaming configurator, but besides that, you really need to know what kind of data the client is offering as an input – and what quality the data is providing. Let’s talk about CAD-files first.

If you are dealing with a small company, you probably can get in touch directly with the person who creates the CAD-files, which is always a good idea, because they can make your life a lot easier. The larger the company, the more departments will be involved, up to the point where every-thing might be stored in a company-wide product-data-management (PDM) system while the creator and the user of engineering data don’t even know each other. In that case you usually send your contact searching for a product manager, which himself should know who to talk to in various departments cascading from top to bottom. Keep these lines of communication in mind, when planning deadlines and writing requests, because inquiries passing through large companies can eat away all your buffer time and the person pushing the “export”-button might never have seen the briefing.

Having found someone willing to help, what do we even want? Similar to the CGI software landscape, there’s a few big software companies dominating the engineering software market and some less known solutions, but luckily there are a few exchange formats that almost any of those software suites can export and that you can handle. For example, STEP and IGES (*.stp and *.igs) are go-to file formats you can always ask for when the client will share 3D-engineering data. You will probably not have CATIA at hand, so ask for a proper export.

Another valuable tip: a commonly used software for data preparation inside companies is Autodesk VRED. If the client wants to send you *.vpb files (VRED Binary), you will be able to open these files only if you have ‘’the big VRED Professional’’, which is as of now only cheap at 1.805 USD monthly (or 14.440 USD annually). There is one “small” version of VRED (VRED Design), but it doesn’t open *vpb. The easy solution here is to ask them to export a *.fbx file from their VRED and share that with you.

Back to the data itself and the first pitfall: Do you want parts or an assembly? That is one question your contact might ask you, so it’s important to understand what he means. Basically, he is talking about what you might refer to as “references”. If a draftsman creates a screw, he will most likely position it in 0/0/0 of his 3D-space for general use and everyone who wants to utilize this screw will position it himself at the needed position of his contraption. You could just copy-paste the screw and the other parts creating one heavy file with your contraption, but an engineer most likely wouldn’t approve of this method. Engineering is an iterative process and handling changes is one of the most important aspects of the job.

What if you find out that a new legislation forbids you from using that exact screw? You would have to manually find, replace and reposition all screws in your file, resulting in tedious labor. That’s why engineers usually create so-called “assemblies”, linking part-files into a meta-file, only containing transformations and other information, but no geometry data itself. Changing the screw will then automatically update all its uses in the assembly. Why does that matter? If your client just dumps you a pile of parts, you might lose all positioning and you might end up re-assembling the data part by part, which is not something you want to do. And you might overlook that he also delivered a tiny file containing all the “missing” information.

If you want to be safe, you should ask for a parts-export from the assembly. That means taking all the linked geometries and information from the assembly and putting it back into one single file with all transformations applied. If the product is too large or too complex for that, your contact will probably suggest to split this into several files, which is alright, because you don’t want to watch your computer taking an eternity, trying to tessellate a 10 GB part file containing a complete combustion engine.

Step 2: Tess-uh-what?

CAD-files are based on NURBS surfaces (unless you received a *.fbx export, which is not NURBS, but triangulated meshes). NURBS are mathematically accurate sur­faces which do not have a set mesh resolution. There are different kinds of NURBS data, but that should suffice for now. You could try to import this CAD data directly to the 3D application of your choice, so to say importing the NURBS surfaces as NURBS, not converting them into triangulated meshes. Many will succeed in doing so, but you will notice pretty fast that most of your workflows and pipelines are made for polygonal data and will fail on these NURBS surfaces.

Converting CAD data from NURBS to polygons is called “tessellation” and is a common step for engineering to CGI pipelines. In theory, many of your 3D applications are able to do it, but most of them deliver bad quality, confusing settings or will slow you down with poor computation or annoying dialogues. The best way to go is to decide for a dedicated tessellation software of your choice like “PiXYZ” to batch tessellate your client’s input and export a fbx-file. There are many more, but we find this provides good quality, speed and is affordable. If you are not planning to invest in such software, you can use Unreal Engine for the job, which delivers decent results and the possibility of *.fbx export. Cinema 4D also has a decent tessellation algorithm. There is not much you need to know about tessellation apart from a few parameters that control the resolution of the output, the most important ones being maximum chord (length) and angle deviation. For both, lower numbers mean a more detailed result.

Step 3: Tailoring the data according to your needs

So just set both to near-zero, right? Not quite. Adjusting the tessellation settings can make a huge difference regarding the poly-count and file size of your project, so these values should be chosen wisely. If you are planning to deliver 14K still images showing product close ups, you of course need pro­per mesh resolution and don’t want your shapes to be cornered and chiselled. But even then, it might be wise to use a mid-resolution setting for most parts and just use higher values on some highlight shapes. Investing some time to find values your artists feel comfortable with is worth it, as excess poly­gons will slow you down heavily later in the process. Try to keep everything as lean as possible.

Talking about excess polygons: When processing engineering data, keep in mind that it’s not specifically designed for visualization purposes. Engineers store a lot of information in there, that might be relevant for manufacturing or crash calculations, but not for you, resulting in additional points, curves or geometries that you do not need. Deleting 400 spheres representing welding points or an additional pre-processed engine block without drilling holes can save you millions of polygons. On top of that it will also make your product depiction correct.

If your project aims for real-time or gaming applications and you need to reduce even more, you have two options, both valid: Go for high resolution and try poly-crunching, retopology or remodelling to create light meshes based on detailed input or you could try to tessellate the data with coarse settings to achieve low-poly meshes directly from tessellation. This can be especially helpful for complex shapes you don’t want to remodel but need to be present in the background or out of focus.

Depending on your input and the tessellation application you are using, you might want to take a look at some things before moving on. It is common in engineering to create simplified, airtight bodies from assemblies for further processing in digital packaging. You might want to call these things placeholders. Engineers need them for accurately crafting the models. They need to know where things are, even if they just get an exact shape, not the detailed
part itself. This is common with OEMs who receive data from various different suppliers. But this means multiple objects are baked into one NURBS object. If you are lucky, the engineers kept the materials from the original objects, enabling you to separate them by material after tessellation either directly in the tessellation software or later in your 3D application. If you are not that lucky, you have to select the objects manually. The worst case is that everything is merged together.

Then the only way is to select and extract surface patches (connected meshes inside the big mesh) from the tessellated mesh. Don’t worry, if you are not familiar with the term “patch”. Tessellation algorithms usually keep the object layout from the CAD input. This means for example to define a cube in CAD, 6 nurbs-surfaces will be stored in one object. After tessellation, you will have a mesh containing 6 surface patches, no matter how many polygons within the patches are contributed by the tessellation settings. But what’s the difference to a cube you can create with your 3D-software? The difference is, that the outer vertices of this “CAD cube” are not welded together, enabling you to easily select and extract the original surface-patch in your production software. You can select them in almost any 3D software by double clicking a triangle. It will expand the selection to the most outer boundary of the mesh, thus selecting the complete surface patch. Maya pro tip: Apply a planar UV projection and then switch to component mode (= select faces/polygons) and then use the SelectMeshUVShell function. Super fast, makes a very unpleasant task almost fun…

If your pipeline requires you to merge vertices at some point, it is recommended to do this after you have finished shading your assets or have product correctness approval from your client, because otherwise you will go back, split and re-import a lot. Merging vertices only improves performance by a very tiny bit. So, it’s better to just keep unmerged vertices. Unless you want to create nice UV layouts. But even then, it’s possible to keep them. Duplicate the mesh, make your UV magic and transfer the result back to the original mesh. The vertices are the same, so that’s possible without any issues.

Let’s talk about shading, because we just touched UVs. By default, CAD-data doesn‘t have any proper UVs and tessellation software usually applies UVs per patch and sometimes even random multiple UV sets for one part. So, in short: The UVs of enginee­ring data are a mess, unless you actively change that.

Working in CGI, you probably know your way around UVs to some degree, but the unique properties of engineering data can make this task quite challenging. There are no nice edge flows you can select for cutting and sewing on triangulated meshes and you probably need to sew a lot of shells first, before you can think about actual unwrapping or fine-tuning. Take your time to devise a method that works for you and your pipeline. Time invested into creating a UV-clean-up-script for your production probably pays off pretty fast. You can, of course, just go with triplanar-mapping for most objects, if your project allows it, but cleaning out unneeded UV-sets helps to optimize your scenes. It won’t speed up or dramatically decrease
the amount of data, but it makes handling the meshes much more comfortable.

Before spending any time on rigging and animations, you should check the hierarchy and object naming of your engineering data. Especially real-time engines are sensitive to scene-tree depth and are likely to crash or slow down when importing engineering data without adjustment. Tessellation algorithms have a tendency of creating long mesh names and numerous layers of hierarchy derived from the CAD file and internal translation which can result in complex parent-child relations that you usually don’t need. Those who have encountered this know how crazy this can get. 100 me­shes and thousands of groups. Unparenting, regrouping and restructuring is advisable, also clustering objects in groups that are suitable for your animation and rendering requirements. Also merging objects toge­ther in a reasonable way helps a lot. But always be careful if you “destroy” the hierarchy. If changes come later, you have to update things manually.

Another thing that can slow you down is the sheer number of objects in your scene. Engineers and designers are not concerned with frames-per-second or amounts of objects. “Somehow feasible” is good enough for them. They even need it, it’s their sophisticated workflow. It has a reason. But this can result in data containing a vast number of small objects and fragments like ball bearings in a gearbox or even single black print dots on a windshield, sometimes CAD-modelled seams and stitching (it’s rare, but it happens). It is advisable to combine as many of those like-shaded objects into one as possible, keeping certain necessary separations for animations and Z-buffering in real-time in mind.

Also, take a look if you imported data types that you don’t need. Engineering data can contain lots of curves and visibility-layers that often survive tessellation and that you in most cases have no use for. Deleting them helps to maintain the readability of your scene and avoid cluttering your GUI with cryptic stuff.

If you still need to reduce file size or require more performance, you can think about deleting unneeded triangles from meshes, trimming them to the visible sur­faces so to say. This is a very time-consuming process when done manually and heavi­ly relies on smart selection methods, like patch selection and sound understanding of angle-dependent visibility. If you do not have that kind of manpower, you could also consider using ray tracing based optimizers that shoot visibility rays from different viewpoints, deleting all triangles that are never touched by a ray. Bottom line: If you have clean data to work with, it will be a joy. If you don’t, it won’t.

Step 4: Candy-crush screws and bolts

Having removed all the junk from tessellation and conversion you now probably ended up with a 3D scene containing bits and pieces coloured in all shades of the rainbow, lacking a certain degree of realistic look and feel. Although some things regarding that topic have changed in the last few years, engineers usually don’t care about realistic shading when working with CAD-files. They focus on good recognizability in assemblies applying as many different colours as they can. But are they completely random?

Usually not. Often you can recognize identical physical materials by their common colour, for example all steel parts being turquoise or cast iron parts being purple. Plain red and yellow however are to be approached with a certain caution, as they are often used not to indicate material, but recent changes on a geometry. Asking for reference images from the customer is probably unavoidable, but taking engineering shading into account can save you a lot of time applying your viz-materials. If you are the curious type or your client fails to obtain any references, you could try taking a look at the CAD files yourself using a free CAD software like FreeCAD investigating the CAD materials. Often the CAD files or accompanying drawings contain detailed descriptions of materials like “St.8.8 Zn-Ni BLACK” that can be easily googled for reference.

If you are planning to create real-time applications, you probably want to merge as many materials as you can without losing too much product correctness. Again, try to combine as many geometries of the same shading as your rigging and animation
allows, as it will greatly improve computation cost.

Step 5: Your actual work

Congratulations, you finally get to do the actual creative work. How hard your journey was until this point depends very much on your client and his data hygiene as well as his ability and willingness to get in touch with the engineering people. The good news is, many of the steps aforementioned can be automated or scripted to some degree, usually doing some python scripting. But it is important to know what’s happening and why it is happening. For some jobs, processing the data won’t be enough and you will have to get involved in soft part modeling. For engineering, a rough representation of a motorbike bench or steering wheel is good enough for their purposes. Sometimes you might only get the inner part on which the upholstery is applied. To CAD realistic fabric stretching, wrinkles and stitching are details which do not matter and that you will have to add manually based on references, 3D curves, drawings or any other information. Ask for every bit of information the client can get on these parts and if possible
request for a reference photo shoot.

If your workflow allows it, try to keep one “master file” representing your single point of truth for the project to avoid local deviating versions. As mentioned before, engineering is also about handling change, and you might have to replace assets in the middle of production due to an update. This helps also to manage the lifecycle of products which might change over time, blessing you with image update requests. Thinking about future projects or pitching, master files are also very helpful as a starting point for cross-platform derivatives. If you have a clean Cinema4D scene set up, data smith import to Unreal is pretty straightforward and reproducible, even after you changed your cinema scene.

Bonus step: You can get visualization data from your client

Sometimes, clients offer to deliver their own visualization data as an input, so that’s the data prep jackpot, right? Again, that strongly depends on the project you have in mind. If the job is to create a couple of fancy still images you will probably be fine and the worst thing that can happen is that you have to recreate shaders for your own render pipeline. Sometimes you get horrible input which you have to touch a lot anyway, but sometimes you get a readily shaded, fully textured master scene in your choice of 3D application. It did happen to us sometimes.

However, if the client wants you to create a real-time-performing dataset, caution is advised and you should probably ask for some sample data before taking the job. But what’s the risk? Especially large companies tend to optimize processes with certain automations and go for approaches that somehow work for most things but don’t always produce clean data. This can mean that they tessellate everything with slightly raised values leading to insanely large polycounts or this can result in some LOD-pipeline delivering nice polycounts but weird topology and hierarchy that can be really nasty to handle. If performance is key, don’t hesitate to ask for the original CAD files, as it might be much easier to re-process the data than trying to get someone else’s highpoly assets to match your purposes.

Summary: If you know why, you will find out how

Everything completely depends on the exact job. Real-time or high resolution still images, animation, or everything in one project. The best practice is to analyze what the deliverables are and how to get there the best way. Not blindly running and preparing data,
noticing later, you’d need a totally different approach. Even if you run into problems and don’t know how to repair a mesh or even get it tessellated in the first place (some broken parts just don’t work), knowing what’s behind all this will help you to overcome these challenges. We hope this was enlightening to you and will help you to work with engineering data of your clients. Godspeed!

About Mondlicht Studios

Mondlicht Studios is a CGI Studio located in the heart of Germany. Boasting a significant background in the automotive sector, their ex­pertise is a testament to their deep roots and understanding of the nuances in this industry. The foundation of the company is built on the rich experience of their co-founders, each bringing close to 20 years of professional expertise to the table.

Surrounded by a small team of veterans and professionals the team is delivering award winning content, providing aid and helping clients to step up. Among these clients are Lucid Motors, LVMH, Nissan, Panerai, BMW, Netflix, Amazon, Netflix and many other renowned brands.

Over the years, they‘ve been honored with numerous awards including Promax Global Excellence and Clio Entertainment Award. The team‘s dedication to excellence has also led them to be named among the top 200 digital artists in 2021, 2022, and 2023 in the Lürzer’s Archive.
Mondlicht Studios has also been celebrated multiple times at the Motion Design Awards
and in 2023 became certified service partners of Epic Games. Beyond their achievements, they believe in giving back to the community. Members of the team have had the honor of serving as judges at renowned platforms such as the NY Advertising Awards and Promax North America.

mondlicht-studios.de

The post Engineering data to CGI –what could possibly go wrong? first appeared on DIGITAL PRODUCTION and was written by Mondlicht Studios.