The Future of Materials in Unreal Engine | GDC 2023
Vložit
- čas přidán 23. 08. 2024
- Watch the recording of our presentation ‘The Future of Materials in Unreal Engine’, originally shown at our booth at GDC 2023.
As we push the boundaries of what’s possible in real-time rendering with Nanite and Lumen, so too must we push forward materials. Learn more: dev.epicgames....
Let’s peel back the layers of Substrate, the upcoming paradigm-shifting material system for Unreal Engine 5-and get a sneak peek of what’s to come!
Check out our blog post for news on UEFN, 5.2, MetaHuman, and more: www.unrealengi...
#UnrealEngine5, #GDC2023, #EpicGames, #GameDevelopersConference, #UE5, #GameDevelopment, #Substrate #Materials
Holy crap, finally an Unreal presenter who was clear, prepared, and concise lol So many of these GDC videos are... not great. I appreciate how hard it is to present this technical stuff, but sometimes I get so bored waiting for them to hit high level concepts. Bravo!
What are you talking about, Epic's UE evangelists and tech presenters are notoriously on top of things with mostly well-thought-out live demonstrations, in itself already a huge challenge. This, PCG, the overview... lmao they've been crushing the presentation game for more than a decade at this stage.
@@minhuang8848A good presentation hits high level topics and concepts, then dives into minutia and technical details. Too many Epic speakers forget to cover high level concepts and dive right into the nitty-gritty tech stuff. Even if you're well-versed in what a speaker is talking about, it's important to tee up your technical talk with simplicity and an overview of the topics and sub-topics to prepare the brain to receive information.
This video did that. Many presentations do not.
Good and clear documentation and example repositories are what we need :)
Downloading samples and experimenting along with a linked (can be a small) tutorial is worth gold to users and beginners!
The UI Material Lab is a great example. Breakdowns of what the functions do all the way up to examples of full working widgets so you can follow how it all comes together.
Nothing worse than Epic's Documentaions. It makes me wanna un-Existance myself
This is very exciting. I really hope to see more upgrades and improvements to the material editor itself as well. Also, something like a built-in reflective / refractive index node with common material static values would be really nice to have. I realize I can easily create this myself, but it seems like something that could come out of the box.
As someone who is coming from GPU Offline Render engines, like Redshift and Octane, this approach to material authoring makes much more sense to me. It’s basically the same thing.
Exactly ! It's a great addition ! I guess I felt a little "taken for a dumbass" with the presentation's emphasis on this "it's hard not to be SO HYPED" paradigm shift... This is Unreal catching up to modern render engines, it's great, but the ground being broken isn't that new (unless i'm missing something only substrate can accomplish ?)
It is the same thing, I've used this in Modo for ages.
@@scottduillet The difference is it's realtime
This is really awesome. The milky substance over the metallic copper really shows what this is capable of. Instead of cheating and using blend masks you can just describe the thickness of each layer and its then physically accurate. Thats amazing. Cant wait to create all of the materials that I previously havent been able to make. This is the best new feature for sure. ❤
I can't explain how happy I am about substrate btw!!! I understand it cause I actually read a pbr guide and I'm so excited to be able to expand outside the limitations we had in the metallic roughness workflow. Especially because I love gemstones, translucent, and iridescent materials.
Where did you get "Stratum" from
@@IPpainting It used to be called strata haha. I seem to not be able to remember the new name so for a second I thought it was stratum. Substrate like a sub material. I'll keep that in mind.
OMG, I thought it will be just a change of some nodes and material models but this?? It is so awesome. I am so looking forward to playing with real physical properties. I remember creating a simple path tracer and I implemented conductivity, etc., and the results we much more life-like. I will definitely use it as soon as it is out of experimental. I hope they will be able to do something similar with volumetric effects.
This presentation had me mouth open and slow clapping on my own. Even though I'm not currently using UE (and I haven't, really), I am so fascinated, impressed with and excited about all of these new features and additions. What a time to be alive for game development!
I am super duper excited for this pipeline. Very cool. BTW, this guy is a great presenter and instructor.
Also nice shoutout to Ben Cloward
I am so glad you finally fixed transmission in Unreal, It was really hard to make decent looking glass previously and was also really expensive to run.
The feature is all I was thinking in a dream. this is truly epic.
Very exciting improvements in physically based rendering :) looks like having all of these helper nodes will make a lot more work for artists though who can achieve similar end result now with just 3 texture samples (albedo, normal, and ORM). Feels like the notion that they were building a system for artists was lost somewhere along the way....if you need to say "so here are the extra steps required to get what you ACTUALLY want" then the system is just ergonomically wrong.
But it is obviously powerful so hopefully some work will go into making some of these functions more implicit (e.g. if you will almost always want to use a resulting color for SSS instead of absorption, then that should be the default input instead of having additional nodes that you would need to add every single time just to do the conversion for you).
Fully agree, if someone wanna do more work, it should be optional by adding new helper nodes rather than adding helper nodes to do less work (which becomes long work as a result of adding extra things to do less work, so there is a paradox there...) They should just make a artistic strata material that automatically do conversion under the hood, so the people who dont wanna use scientific values or more textures or more extra nodes making crowded, they can just use that artistic node that automates it and allow for old workflow.
Yeah I'm thinking the same thing. When we need to take extra steps for the simpler option, we need to reexamine the workflow. The other thing I think about is how cluttered and clunky making a stylized game would be with this system.
@Unreal Engine It would be good if you gave us the link for the page where we could see all the numbers for different metals,or non metals i guess.
Hey I remember I learned about the wave nature of light (I know light isn't exactly a wave but yeah) in year 12 and I always used to try the light double slit experiment in different render engines by using path tracing and it never worked. Back then I didn't know how path tracing, textures and etc functioned in a way to be optimized. After researching about it, I thought it was going to be impossible to make light even interfere with one another to create that constructive and deconstructive effect in current computers but now I am really impressed by what you mentioned at 18:04 and I can't wait to see the results for myself.
Awesome presentation, awesome feature, clear examples. I just start to learn UE and try to keep up with everything that happens around it, Epic is really on the edge of technology
lmfao "I don't know about you guys but I spent a lot of the early days of the pandemic watching youtube videos about tempering steel" RIGHT THERE WITH YA BUDDY RIGHT THERE WITH YA
Haven't been so excited for a material change since PBR, well done guys (now give me back my additive decal blending)
This seems very fun to play with, as you can use the properties of real life as a reference rather than just a visual reference.
This actually makes more sense than PBR
Thank you for share this.
Anyone know where he left the reference to the reflective indices that he said he'd provide at 5:26?
I am looking for this myself and the QR code at the end of the video does not have the link from what I see
Honestly PBR made life so easy for me. But As long as the decades of PBR maps and materials will still function then I'll learn and trust this for the better.
With love From Nigeria
so if i'm using substance painter to author materials, would I just export different masks and the normals for this workflow?
Love your work😄
Looks like Engineers at Epic are watching what Blender has been doing :)
and I like it
Yet another new feature in UE 5.2 that reminds me a lot of Blender, this one is very similar to how Cycles nodes work, and the other one was the Procedural Content Generation graphs that are very similar to Geometry Nodes. Looks like the UE devs are getting inspiration from Blender, which is great.
Both Blender and UE5 adopted procedural approach from Houdini
@@VRchitecture cool! I didn't know that
17:07 that actually brought a tear to my eye. Fantastic xD
Super cool stuff!
Materials have never made sense to me, but this seems like it just might. Also based on my tests 5.2 looks like it might be pretty stable. I will consider mastering this "Subtrate"
The heating up engine and soap bubble. Wow. PHEW the frosted glass though!! Exposing these physical parameters for us to directly control... this is insane. I want to try this out RIGHT NOW.
You can
Question: Would it be possible with Substrate to calculate edge wear without the need to use special textures and unwrapped models? Example: Worn edges on a gun, etc. -- Also using the AO/Cavity of a mesh and baking it down directly into Substrate for further use? (not Post Process). I am investigating how to optimize the use of Nanite meshes, since they are hard to unwrap occasionally.
Speaking of Post Process materials… How will this work with Substrate?
How is it possible to make edge wear without baking textures? Substance painter does require baked textures for that, for example. I dont know how it is possible without post process gbuffers
Awesome, can’t wait 😎
I hope this ends up being more flexible than their current shaders, I've been a bit frustrated by my inability to create certain effects and match certain material types in Unreal so far. Especially for translucent organic substances. A lot of these previews seem focused around glasses and metals though, which is cool, but I'd like to see a wider range of examples.
You have the Principal BSDF from Blender now, that's actually super pro and I'm glad to see it simplified for Unreal! I would honestly love a more Blender related workflow for materials as a whole due to the wide amount of mapping and blending options available.
QUESTION - does the Material Count 1 bsdf, 2 bsdf, 3 bsdf viewmode represent material draw call as well?? Does that mean draw call on a material is now on a per pixel basis???? like nanite?
this is crazy exciting!! WOW!!
Thanks.
Öztalay kardeşimize de selam olsun :)
Merhaba!
@@mattoztalay5713 Merhaba :)
Is certainly cool to see Unreal Engine reach parity with off line renders but for those that don't know, this is only "ground breaking", in game engines, this material system has been available on off line renders for decades, for example in montecarlo raytracing engines like YafaRay render and on the renders of 3D tools like Blender, 3D Max, Maya, C4D, Modo, etc.
feels like im back to maxwell :)
This kind of material definition was really expensive, infact it was used in realistic unbiased rendering engines like Thea Render and a few others. Still I doubt this could run easily in the near future for mobile development.
Amazing, but hey, when we will see back tessellation or parallax occlusion?
nice, metallic always was a problem. this looks like the right way :D
Does this remove the Fresnel node :(? Shoutout to Augustin-Jean Fresnel btw.
Has anyone been able to find those reflective indexes values?
Finaly a sensible default. why the standard shader doesn't use the scientifically by default like this in most systems is just beyond me.
Will strata be able to be converted into mobile shaders at compilation time?
Amazing
Looks nice. Does this work with a deferred renderer? It shouldn't be possible to fit all the slabs info into a GBuffer without making it absurdly fat.
This is awesome.
It already reached rocket science. :)
So this the new system is all about "physical surfaces"? But what about all stylized materials? Old system was cool because it doesn't limit to the real world surfaces. New system doesn't allow me to do cool things like toonshading and stuff? I mean old system didn't allow it too, but I hoped the new one would fix it. I want to actually PROGRAMM my materials. Programm final pixel, not bunch of input values - gbuffers or that bsdf stuff. I still can't create my own shading model, right? I have to use that "physical bidirectional surface thing"? I don't get it. There's still no control over color of the final pixel? Because that's what it is in the end. All this PBR, strata, buffers, values, light - in the end it needs to give just a pixel of a certain color. Can I have actual full control over it? Can I write my own shading model/BSDF? No?
This might be somewhat better since it's not as limited as the old system. I know the old material system was actually pretty bad for cel shading.
great job. i'm pleased. substrate will become default going forward. i need a book with a full in depth education and analysis on this subject. away with multiple textures to achieve this.
this is exciting
How can I eliminate the unsightly black reflections on the bottom of the metal sphere? 5:34
Does substrate only really build on surface materials or are there changes coming for post process?
Love you Unreal
As a software developer I would probably do anything to be able to work on such project, so I understand why it's here. As an UE user, I sadly don't feel the benefit yet. If I would want to have nice rocket engine shader, I would do some research, gather reference images and notice, that it's getting some nice rainbow'ish colors in hot areas and I would just animate that in my material. I don't see the benefit in having to learn how this happens on a physical level to be able to accurately reproduce it. On the other hand layering slabs sounds like an interesting design and will most likely allow for very visually interesting shaders to be created.
The benefit? So we can soon be unable distinguish video game from reality ;)
You're missing the point. That's just an example to demonstrate the physically accurate BSDF through substrate. the point is that it now has almost feature parity with offline renderers in terms of standard surface/principled shaders.
@@blakeXYZ We could already push toward that without Overhauling the whole Material model.
i'm in a similar vibe, I actually need some concept materials that i coudn't even approach without Substrate to be convinced of the purpose. A greater ability to copy reality is neat but its not (yet?) a very exiting nor cheaper approach.
@@lasereyes5720 Can we already use real world physical properties, e.g. reflective index of gold? Genuine question.
He mentions moving away from abstractions such as Specular & Metallic and using real world units.
It does look awesome but for me the biggest issue is that artists are not physicists, so this model seems to be adding another barrier for artists instead of removing them. The PBR model, although very technical in many aspects, can give artists a lot of artistic control. What artists need are dumbed down technical requirements so they can focus on the final artistic result. This model doesn't really do that. You might need less nodes to get a realistic result, but you need a lot of physics knowledge to understand how to stylize and art direct something as an artist. What I'd expect to be very useful is a model that lets me customize the final looks in a straightforward way, including for stylized applications. We need to remember that ARTISTS are the main users of the shader system and texture authoring, so the way forward should be providing an interface that artists can easily understand and have maximum creative agency.
In the talk he literally says we can input any value in in any way we like, their goal is to make a consistent plausible result by modeling what happens in the real world not some basic abstractions that mostly work. Your not going to get that with the old/current system. Yes it was somewhat more intuitive to people with "big dumb art brain". But it had a lot of limits. Like what was mentioned in the talk where metalness blending was basically impossible.
I have a hunch that the more art people are "forced" to use the new system the easier it will be. Like with anything else. And if they can't after years of trying they probably aren't cut out for it. Tech is always evolving
Epics goal since idk probably unreal 1 was to make the most realistic engine possible. And you can really see that in ue5 with their demos/tech. This is just another step in that direction. They have less priority on other styles but doesn't mean that they can't be done. Just I a some what different way.
According to other comments here that I've read this new system is pretty similar if not the same as offline renderers. You know the ones that can get realistic enough gci for movies. So this isn't an entirely new concept. Some artists won't really have an issue trying this out.
There is also a legacy converter node as well so you could theoretically just use that to get your old ways to work with the new system.
I don't know ... I'm worrying about those 4000 instructions in the fragment pass itself! ... The cutting-edge tech should cut edges in my humble opinion!
Well (table flip) lets re-learn everything again.
And I was just getting my head around PBR in substance😅 but honestly this sounds awesome, and more scientific (and exacting) way to describe light behaviour in terms of how it affects how we perceive materiality. I think some presets for typical materials will be welcome, so we can study how each node is used.
Phenomenal
I assume it will work with Path Tracer in the future as well?
Does anybody have the list of links he's talking about?
That's the one Marv, that's the silver Tuna..
This looks really cool. But seems like a weird direction to go, when offline renderers are moving AWAY from this, and in stead are simplifying a la Standard Surface and Material X.
Wow, like in Blender!
👋🏼🇨🇴🧔🏻👍🏼🤝🏻🕹️Buena conferencia
How he makes this bubble material?
And the instruction and sampler counts dont matter anymore?
they never have. technically they aren't even all that accurate. Ben Cloward here on YT has a couple videos (Shader Performance Measurement) on that if you were curious
its time for ue5 to make materials compatible with substances
so if this material system is using bsdf... I wonder how compatible with blenders bsdf materials it might be
Hopefully RTX 4090 will be able to show the beauty.
🤩
What i expected and wanted : Easier way to make actual multi-layered material
What i got : A specularish workflow to make very-niche effects that may be more accurate but will cost more in term of performance.
I feel like its not really worth it (yet?)
Please provide the car shader so we can actually dissect it.
It’s like you didn’t actually watch the video…weird
@@tony6795
Not only did i watch the video, but i tested substrate since release, as we have several nodes (slabs) per shading model and multiple operand nodes its really not Easier, the team tell us it is but fall short to showing it, now if you ever worked within the material editor you'd have noticed the specular workflow make it easier to 'copy' material from a data source not to 'make' materials, in regard of making nothing beat classic metalic pbr, especially because of a single color map then multiples attributes masks on a per-need basys (you can then mix in a single map). Now if we saw how the rocket-engine material slabs were stacked and his 'heat' parameter we could have appreciated how the tech help , because i could construct a material that does that if i add 2 (extra) gradients but i sure dont know how much actual heat ever affect the metal, The bubble thin film would have been somewhat impressive if he animated said thin film thickness live with a lerped mask.
Some of the changes like Roughness frosting the class are welcome but could have been added to the previous model without a need for a 'paradigm' shift.
@@TheDesknight hold on, I’ll be right back
Epic is making the matrix
Ah, so that's what f90 means!
Are you guys ever releasing proper documentation for this? I've been bashing my head against the new system for weeks now and I can't tell if I'm too stupid to make something work or if it's buggy because of experimental state of the system. It's incredibly frustrating that you're releasing these without proper documentation to accompany them.
5:30 now I represent the actual physical properties of gold: doesn't look anything like gold (╯°□°)╯︵ ┻━┻
Epic devs really should take their time and look at some actual CGI renderers (Vray, Octane, Maxwell etc.)
because for now their "paradigm shifting framework" is just a complete mess of terms and nodes nobody uses. basically Substrate does exactly the same as any other material system but with every single element having weird name.
where is glossiness? where is IOR? how a CGI-artists would convert their work to a Substrate ?
it's like someone gives you a "paradigm shifting way of movement" which is just a regular car like any other, but you steer with a pedals and shift with a steering wheel which also controls wipers (because it's a new convenient pipeline)
I get that feeling too. As if for marketing and PR reasons there is attempt by various companies to name same thing differently and claim its something new not just for their product, but overall its some new GAMECHANGER thing. But if you scratch the proverbial surface, 9 out of 10 times its juts PR teams indulging themselves.
When you add the coloramp Vector curves Adequate noise And nodes for working with color correction???
The presentation is great, but, the disparition of Metalness, roughness and oppacity create extra gaps between technicians and future artists ;
When teaching art-making and optimisation in Unreal 'F0' and 'F90' mean nothing, Maybe a simple option to rename attributes (that doesn't ship) would fix it,
The Metalic PBR model as a base was better , I'm sure you could have cracked the dialetic to non-dialetic barrier in the code without an overhaul off the complete concept.
Because i'm pretty sure in this specular oriented model if my material has colored metal and non-metals, i now have to bring 2 color maps per material, wich is fine for movie makers but double that cost will compound very quickly at 120 frame per second.
We loose a lot of accesibility in the name of some potential effect exactitude that will only be usefull in niche cases, like thin film 'renders'
It does convert a lot of the guess-work (that was intuitive once you get used to it) into reshearch but we will have to reshearch material values (kind of a chore) AND reshearch what all the new nodes do , so please :
At least provide the complete written documentation per version of substrate (if there is 30 News nodes i need 30 detailled descriptions with examples, because current subtrate is hell to navigate , the UI is no issue but we need the new obvious ' correct' way to do things , before we can try to go around it to optimise.
TLDR : i tested it and am not convinced (quite the opposite) Please bless us with the User documentaion + a complete project with 2-3 incredible materials only Substrate could manifest, and 2-3 generic materials built in a way that is not much (ressource) expensive than it was vanilla PBR.
Yeah they wont do that. Also, take note, he said three times at the start: "THIS IS WHERE WE ARE GOING". They dont care if you have problems; this is WHERE THEY ARE GOING. To me, the system is a backwards step; even just plain old Disney BSDF was easier than the mess of extra work they made here.
I agree, sad to say but I didnt like it.... Unreal engine material system wasnt so well , and now it seems worse with this. Its bad to add new nodes just to simplfy things, it should be opposite, if someone wanna complicate things they should add new nodes to plug scientific values, not the opposite where we plug things to simplfy things.
@@Joviex Actually cooking a video to help beginner with it , The metalic workflow is better, simpler to explain and memory lighter, knowing that we can guess that The global update is not targeting video-game making (that' s kind of a sin). we'll always have the UE4Disney node (Substrate compatible) . Overall i agree its a step backward, because it make possible to make very cool edge cases materials, but the creation of basic material is now more complex. With exeption for the glass routhness blur the whole substrate should have been a separate model like hairs and surface water.
@@mcan-piano4718 Correct, its making material creation more like you would ' outside the engine ' (in offline rendering) and that's lame. Maybe Its an attempt to be unique again since other engine are finalyl catching-up with the PBR pipeline, but it sure isn't helping my personal projects or my art-teaching workflow. I made a rambling video explaining it to people who already have a basys in materials and it is still only grazing the essentials after 40+ minutes.
This is crazy. Amazing how you can increase graphical quality and performance by modeling real life.
Not performance. If you noticed some of the instruction counts were in the thousands. Where typical mats are in the low hundresds
All I want is photorealistic games on PSVR 2.
we still need normal mirror reflection in lumen... still wait !
What do you mean?
Very nice, but really need dumbed-down abstractions to make things simplier
now ,i should Bcs in material sincse.
awesome! but now we need artist friendly naming and usage for this to be perfect :D
18:21 nope, I disagree :)
It also feels like the substrate doesnt achieve too much of a difference, it just complicates things, while with PBR hyperrealism was very much possible.
Perhaps its best for people with a physics degree and/or tech artists.
We'll see how it goes, but Im very skeptical so far due to its complexity and showing something I can do with PBR anyways. Performance and stability is whats important imo.
now another thing to learn .
How about a node that comes with all the accurate metal values? Now you still have to write down all the correct vallues, but with this idea for a node all the excisting metals will be on there. If you want copper you just connect it to the copper socket.
Soooo.... standard PBR Shading a-la most offline renderers since 200x?
a brand new Pa.Ra.Digm Shift ! (Only if you've never used any other modern render engine, ever)
Unreal Engine will become UnUnreal Engine
😮
Full Path Tracing in CB2077 is pretty great.
The future should be LLM + Unreal engine.
We need tessellation back
There's no difference in the end result. The only difference is users can now find an established value online and input it, rather than having to fine tune the values directly for lack of proper word filtering.
They're preparing the engine to have an "Ai content generator algorithm". The new word filtering will allow the Ai to get values online and apply them directly without having to understand which metallic and specular values are equivalent to established science given data.
and what would be the problem about that?
@@detuneCris You're assuming there's a specific emotion driving the comment but there's none implied. Reply based on the given prompt 👍
@@n1lknarf problems can have an emotion behind them or not, but I'll allow it.
Let me rephrase that in a more simple way that would hopefully be easy to understand:
"So?".
Meh. What I would appreate if 3d Software like UE and Blender would ship with 100 most common, ready-to-use procedural materials, that you can customize with some sliders easily.