Compression or not, this isn't the most impressive UE5 demo out there simply because the quixel team didn't use 8k assets like they did with lumen in the land of nanite. You can squeeze much more out of the tech in realtime, especially on high end rigs and even consoles with optimization and procedural tool like houdini.
I really dig how much effort is put into developing the tools. You can tell that the CEO is in fact a programmer at heart. I hope if he leaves that they hire within the development team of the engine. They don't need to pull a Unity.
One of the main advantages of Unreal Engine is that they actually develop the engine along side their flagship: fortnite. Unity is an engine that's made without any specific games, so it's not really being made to fix specific developer problems, they're just adding random tools and hoping for the best.
8:46 The quality is so bad because the video encoder youtube uses cannot encode that much detail per frame. This usually happens when you have a lot of small things moving about at the same time.
At 3:06 so true! Because the main difference between a game made by a computer school student and an AAA firm is (apart the optimization) is mainly the nicer assets for looking good :) Remind me my OpenGL 3d game as 1st year student project in 2001 running at 10 FPS with 100 awful flat textures 128x128 (while taken from Unreal Tournaments), no lights (too many GPU perf) and my 1000 triangles car model homemade in 3D Studio Max :D
I visited UDS (Norrköping back then, they was later bought up and moved) back in 1999 and they made a racing game and everything was made with objects on splines. So kind of a halfway house. Also the scenery. Like a palm they had that was a spline, so they could bend it down and away from the track
great insights! always love your views on game engines and development I think it'd be really cool to hear your thoughts about Verse, the new open scripting language by epic. They seem to have very big plans for it as it's supposed to be the driving force for the metaverse, so entirely built with concurrency asynchronous functions and networking in mind
A small explanation. This system is working based on base PCG class, which provides a lot of tools and ways to determine how to shape and drive logic within its guts. It is very similar to Blueprint geometry tools we know from 5.1. @20:56 Those categories can be expanded by subclasses, so if you have a base class which contains this category "PCG", all variables you set to be in this category will display there as well. This way you can easily track specific groups of attributes.
@@astroid-ws4py yes and no. No in a way that procedural nodes are just set of instructions, this blueprint logic can inherit from parent classes. You can create PCG procedural rock formation, child class procedural rock formation with moss and then it's child class procedural rock formation with moss and plants. You can enrich the logic you have from parent class.
I used to design/create race tracks back in the day.. We'd get the actual raceline playable and as close as complete, before building any of the other mesh - basically prototype it. It was the rise of modders + track editors that led to the more procedural development.
I have also watched or read something about foliage rendering already was a part of unreal engine 5, so youre probably right about it coming out in 5.1
Maybe you could look at Counter Strike 2 and specifically the new and improved graphics? It's obvious that there are a lot of lighting improvements but it would be interesting to hear you break down exactly what we are seeing when we see "lighting improvements".
Procedural generation uses assets that are hand crafted, no man’s sky works in the same fashion. The procedural is the way you code or programming tells the assets to work together to form your world.
Have you tried layering a decal material on top of glass? I'm trying to add a top decal layer made with substance but the problem is that the output node must be set to TranslucentGreyTrasmittance for glass to work. So it won't work... the decal is translucent. It would be good to know how did they manage to add those mud decals on top of the opal material.
8:57 The quality is so bad because of the CZcams compression algorithm, that screen was pretty “noise-like” which is very hard to compress, it has a high entropy so the quality for a given bit-rate drops
I think the black pixels in the nanite debug view is just a basic postprocess depth-difference outline shader just to make visualization easier, otherwise it's kind of just a multicolored soup of pixels
Typically when making a large scale physics engine, there are a interaction box that typically is longer in the direction that anything is moving. So if two boxes are intersected. There is a close box for that is right tight to the true physics model. And when the inner boxes touch they use the physics model. And typically artist designed. Some games just use a LOD mesh for it
@@W1ngSMC That us true in some cases, The benefit of a box is that it can be streched in the direction of travle and also have a more exact intersection point. The cost of the intersection points are much lower than the cost of the physics calculations
you gotta keep in mind that these demos aren't optimized at all like a game would. and it's running in editor instead of a packaged build. Dev machines for this level of quality are always very beefy because by the time a project like this releases the hardware and software (like dlss 3) will have caught up just a few years ago this would have had to run on an entire server farm way beyond the cost of a top tier gaming rig.
Yes, but for a workstation it's probably mid at best. They want to point out that their non-baked demo running in editor can run on consumer level hardware with that fidelity.
"Why is this quality so bad" it's the encoding. It can't preserve a good visual fidelity for a given bitrate when the image is noisy. Nanite triangle visualization always does this, as will any kind of noise.
It wasn't just a comment on that particular section, I think. The video was captured at 1080p, making the demo look blurry, instead of higher resolution, like 4K.
For continuous professional development points we call rolling through CZcams Attending A Seminar. If you leave a comment you attended a half day workshop.
Isn't the concept of having to draw 'triangles' a bit limiting, especially if you can define a planar face with multiple vertices? It seems like the 'everything has to be broken down to a triangle' is a limitation of current GPU's ... wonder if anyone has considered the performance benefit of pure hardware rendering of planar polygons?
one point to consider is that when everything is a triangle the way to handle everything is the same (for parts of the process at least), which might lead to simplified special-purpose hardware as well as more easily parallelized works loads
Planes/Quads have to be divided into triangles at the gpu level because the triangle is the simplest enclosed 2d shape. Then you can make gpus to maximise efficiency on doing the same task at the simplest level multiple times at a fast rate aka rendering billions of triangles at the same time across the many gpu cores. Its easier to do 1+1+1 many times with many students to calculate 1558693 x 27343083432 than making them do 358x6933 many times.
@@rudrasingh6354 like, I agree with you at the hardware level, but I'm pretty sure doing multiple "358x6933"s would be faster than essentially counting to 4x10^16, like, that multiplication plus a sum of the result should takes less than the time to count to two million or so : p
They just say polygons so people can understand it better (marketing to n00bz), triangles might be too exotic for some :D The black foliage is probably placed as static model by the artist, as a touch up.
The imported model will likely be done using quads, and Nanite directly processing those quads. Calling it polygons is more precise. The final triangles for rendering are just the result then.
@@vast634 The model itself can be created with quads of course, in the DCC tool of choice, but I do not think it is exported as quads. The model's triangles are grouped by hierarchical clusters (using the METIS library), you dont need no quads for that.
Hey Cherno, can you review the Hellblade 2 new cinematic trailer as well ? It shows off Senua's facial expressions and what they've achieved with UE5 on Xbox series X.
So true, your having fun creating game systems and you remember you haven't even start making the assets that you're thinking of then the dilemma of Artstyle and rendering 😆
8:10 Whether the truck was 70mil tris or polys, he's likely referring to the source asset, and nanite makes it lower poly as needed, so it wasn't rendering many polys for that model as you can see large tris in nanite view.
im 6 months new with unreal 5 , its RVT (runtime virtual texturing ) isnt it ? blending into the surface , they were vague as far as the procedural side , so we, ll have to wait for the blueprint for said procedural , & then swop out for our own asset elements inside blue print. is this right ? i downloaded unreal 5.2 peview , yesterday took half a day ! must be somthing new to view inside starter content !
i still fear the future where you pay 300$ for a game and 200$ is for the 1tb external ssd that is completely full with just one game. we won't have to buy dvds or so but external ssd/hdd to play one game.
I do think we go the other way. One reason games have grewn is banking. Also real world scans. Pseudo real time lighting solve the first problem, and presedual map generation solved the rest. On top of that back in the day (90s) games could be installed with diffrent levels of texture quality. Back then textures was 90-95% of installed game data so if you cut the texture from very high to high the install may go down from 600MB to about 200. Go down to medium and it was under 100. I can se the same happen with nanite, while there is no global LOD there is a sort of LOD equivalent. You don't want a 1.5TB install. Well go down from extreme to very high, and its 400GB.. Still to much, go down to high and its 100GB While that is not a thing yet, I see nothing that would stop it.
@@matsv201 I'm not so much concerned about the textures themselves but the poli count per object. if they have millions of poligons for each single variation of small stones in the scenes and each object if needed or not has the same or more then textures won't be what drives up the storage space anymore. sure we don't need LOD models anymore but we are still stuck with the highest detailed model and we don't have the 1-3 million poligon per level barrier anymore before it slows down to spf and not fps. i wonder how many Gb that shown scene would take after being compressed for storage and how many Gb Ram it takes to run it. 32 Gb of free ram probably won't be enough.
@@dovos8572 To my understanding it works like this. Every object have a standard low poly wireframe model. Same as other game, that would be similar to a LOD-0 model. On the LOD-0 there is a higher resolution displacement modifier with a 2 dimensional matrix table. I don´t know if its a 1, 2 or 3 channel table, my guess its that its 2 channel. Normally a wire frame model need to be at least 16, possibly 32bit to have sufficient precision (professional system sometimes also use 64 bits). But if the level under is corrected by the level on top quantification errors, that would be a problem if it was directly addressed. So Nanite can probobly get a way with 8 bit precision with no problem, actually i think it can get away with 4. (the same method is used in the original MPEG codec, so its really not something new). This would be stored in a file with it segmented in different areas. So the deeper information is stored later in the asset. Here is the kicker.. well there is actually two. A matrix grid is to the computer identical to a texture. So the computer handle it as if it where a texture. Actually the vertex data can be mixed in with the texture data as a extra Chanel. There is a few more tricks that can be made. Normally textures are stored RGB and sometimes with a few more channels like alfa or something. But image data can also be stored as YC channel. This is how oldschool TV compresses color information. The reason why it have not been used in gaming prior is due to lack of mixed precision GPU. But now its pretty common, Nintendo Swich have it, and pretty much everything newer. A HDR YC texture could be 16-8-4-4-4-4. Then you would need a additional 8-8 bit for the model matrix. That would give a pixel depth of 56 bit, while that might sound like a lot, that is lower than say a 16 bit HDR RBGI pixel of 64 bit. So your 3 million poly object would be 3M*56bit... 21MB.. yes.. MB.. not GB. The second kicker. Direct storage. Typically less than 1MB of the object would be in RAM. When you get close to it, the GPU read in the rest of the file from the SSD with direct storage. The thing is, you will never see both the backside and the front side at the same time, and really you would probably not even the side in full resolution. So the Direct storage would read in about 7 or 8MB of it when you are totally up close to it. There is still one more kicker. Texture compression. With placing the bits a bit smart, you could fool the GPU that everything is a texture, so then you can use texture compression. Both lossless on the polygon position, and lossly could be used on pretty much all of the other channels. (in theory you could use lossly compression on poligon-possisons as well, but then it get kind of complicated). That would shrink the object on file down to anything from a few MB up to a bit over 10. And also, a stone or a pebel would not be 3 million polygons. There is one more trick that can be used. Because every level of detailed is stored in order. You can pretty much cut the file at any level of detail and just leave it. It would still work.
Hey Cherno... if AI can now generate image based on their fragments, why not generate game worlds based on small fragments generated by artists or even text prompts. If such a world could be generated in minutes, why not generate personalized worlds for players, based on what they would like to see.. perhaps this is the future of gaming
That sounds like an mindless ideas.You probally can create entire high grade game using ur.engine with its visual scripting for such a while already.The point is not about convenience,but versatility.Or more importantly,what makes us do art?When AI now can also create art to compete with us?Well,art is not about competition,it is in fact about emotion and love and meanings that we as human beings want to deliver to audiences.So let a machine,even a sophisticated one to do the entire work is less of ideal.
You know what I didn't even notice that part you were talking about about how basically this generative level design is probably a precursor towards incorporating AI because I mean unreal has all the quickso bridge and everything built in and if you had a chat GPT4 and somebody's already working on something experimental you could just have the AI organized and edit your assets your art assets you could basically make a game in no times flat because the truth is this you know most creatives just want to get down to creating you know eliminating all the issues of codeine and all this stuff it'll just make it easier for you to focus on and fine-tune your creative idea you know.
Wasn't nanite supposed to not work with deformation and animation? The whole truck is Nanite, going by that polysoup view, yet those wheels are deforming based on the pressure from the ground and gravel.
@@user-sl6gn1ss8p iirc, they said they deform, will look for timestamp, but check out that part where they change the material of the car to that iridescent one and as the camera rotates, check out the front left tire that's rotated. It looks slightly compressed on its bottom side(9:59). Also check out when he tries to move, the same wheel gives the feeling its compressed as it turns forward, so the truck can move on.
Not sure what he says specifically, can't make it out, but in 7:03, he says "... simulates how the tires compress into form". He follows it up with Metasound but not sure if that's what he meant.
@@npatch oh, yeah, fair enough, I think the frames I stepped just didn't show the effect too much. Don't know much on nanite though, so hopefully someone can answer your question : )
@@user-sl6gn1ss8p Me neither, or at least not sure if sth changed since the first reveal, but I distinctly remember it was only used for static geometry, which is why characters, which both deform and animate, do not share that same treatment.
It is the same as AI generated images,impressive but you can guess,AI is so powerful now,it could even create real time animation from streams of video footages.
Wish you would watch it before or maybe wait until a segment is through to talk about it... several times you ramble on and on about something he literally explains right after you resume the video. Most of what you said is immediately worthless, if you would've just waited to comment on it after he demonstrated it.
That procedural editing stuff and all the megascans makes me worry that everything make with UE is going to all look the same - unless you put lots of work into tweaking it. It'll be like "oh, that looks like it was made in Unreal".
That would only happen, if everyone also just uses the same assets (which probably a LOT will do, because they just want use the hyperrealistic Quixel assets, instead of more stylized or low poly stuff). What this whole procedural generation reminds me, is Wave function collapse, or autotiling from some game editors for tilemaps. So while the underlying rules to create a world might always be very similar, the look will drastically change with the use if different assets.
There isn't any obligation to use any of these tools, most teams probably will use them more in backgrounds and the like at most and use actual custom assets in the more focused foreground, or maybe not use them at all so they can get a full style. While I definitely do agree a lot of developers just end up using these assets everywhere they can, I notice it's mostly newer/less experienced ones that just chose unreal engine because it's "the realistic next-gen engine", without regards for other aspects. And then in terms of many procedural things, as mentioned by the other person swapping for custom assets and setting up different procedural rules can have major effects on the style. But of course, there is almost always an underlying look in almost every game engine I have ever seen, regardless of assets used.
In some future when engines can create perfect realism, it all HAS to look the same, as it does so in the real world. (Talking about a scene in a real world biome / same Region). There is not artistic Interpretation in nature. Of course artistic touches will always be possible. Just as a nature scene in a movie will look different than one in a documentary.
I hope Nanite brings that much energy when running on a 4010 because the way Nvidia is pricing their GPUs, a 4010 is the only thing most people are going to be able to afford.
13:55 Yeah, github Co-pilot is already going to have voice control to guide your AI assistant. It won't be long before every tool has voice control. It is honestly insane that computers understand natural language.
it's nice to have but nothing really new, studios build such procedual generations since years in their own engines or even in Unreal itself. I hope 5.2 offers more
Should unreal engine divide video game industries? if people saying unreal engine photorealistic games ruins video games. The truth is I don't know what to say. I support Unreal Engine developers, photorealistic artists, photorealistic game artists, photorealistic game developers, etc. Is it easy to make photorealistic game? yes, quixel, photo 3d modeling, photo realistic assets saves people's time. I don't know why some people are against photorealistic game. I'm not against it. I see bob ross's painting. but beautiful and realistic.
@@marktime7692 It's not the individual car that's the problem, it's the ongoing and unnecessary promotion of inefficient, dangerous vehicles (SUVs, I mean) during a climate crisis. I'd just rather they found a different industry to sponsor.
Outstanding work by the artists and your statement that the engine won't be worth s^^^ without great assets is indisputable. But when AI companies that create art generators gets their hands on these models....what will happen to the industry, most importantly, the people in it?
I really think AI and procedural generation is something that should be frowned upon and I agree with Valve for removing games that have AI art in them. Games are supposed to be worlds that are created completely out of the artist/game designers minds. Using AI will quickly destroy the character and soul of games. If that is truly the future then we should really cherish the media and games that we have now.
AI is a blessing for indie/solo devs, nothing to be scared of, it's just a tool for artists/devs/programmers, if your entire product is made by AI ofc it will be soulless and generic.
Man, one serious downside of these reaction videos is when you're like "yeah idk, he didn't really explain it" only to then continue on with the video where he explains it. Is there any real reason to do a "first watch" video rather than watching it beforehand and being able to give a more well thought out commentary on it? It's not like you're having any crazy reactions to seeing things for the first time that couldn't be replicated on a second watchthrough.
The new Metahuman animation system was also quite impressive
Aha that was amazing
yup, tho no nearly as impressive as the human model system unity is baking. Look it up, you literally cant tell wether is a photo/video or not.
I second this
It's nice seeing someone who actually knows what they're talking about.
his breakdowns are the best
Hahaha true
8:48 Compression, the video has millions of moving traingles with very high contrast between them
1080p60 is like 9 Mbps in bitrate which is wayyy to low for that scene
Compression or not, this isn't the most impressive UE5 demo out there simply because the quixel team didn't use 8k assets like they did with lumen in the land of nanite. You can squeeze much more out of the tech in realtime, especially on high end rigs and even consoles with optimization and procedural tool like houdini.
I was at GDC and an Unreal guy demoed this to me, I must say it was very impressive, the foliage and lighting look unbelievable, quasi photorealistic.
I really dig how much effort is put into developing the tools. You can tell that the CEO is in fact a programmer at heart. I hope if he leaves that they hire within the development team of the engine. They don't need to pull a Unity.
Many good YT videos about the story of UE, from my memory, many good lessons to learn from them
What happened with unity. I'm ignorant
@@N1rOx Dying! They are trying to add a poor idea: coding your game with IA
One of the main advantages of Unreal Engine is that they actually develop the engine along side their flagship: fortnite.
Unity is an engine that's made without any specific games, so it's not really being made to fix specific developer problems, they're just adding random tools and hoping for the best.
@@quentinquadrat9389 "IA"? do you mean "AI" or is this smth completely different
There was a whole talk about the PCG later. It's very nice, you may want to have a look into that as well.
is there a link anywhere?
@@fallohide7418 Yes, there is.
8:46
The quality is so bad because the video encoder youtube uses cannot encode that much detail per frame. This usually happens when you have a lot of small things moving about at the same time.
I loved their demonstrations. Super exciting to see. I'm also blown away by FAB--the new combined asset libraries that just drag and drop. Huge.
At 3:06 so true! Because the main difference between a game made by a computer school student and an AAA firm is (apart the optimization) is mainly the nicer assets for looking good :) Remind me my OpenGL 3d game as 1st year student project in 2001 running at 10 FPS with 100 awful flat textures 128x128 (while taken from Unreal Tournaments), no lights (too many GPU perf) and my 1000 triangles car model homemade in 3D Studio Max :D
i love this content from you! Please do more when you have the time. Thanks! Cheers!
I visited UDS (Norrköping back then, they was later bought up and moved) back in 1999 and they made a racing game and everything was made with objects on splines. So kind of a halfway house.
Also the scenery. Like a palm they had that was a spline, so they could bend it down and away from the track
great insights! always love your views on game engines and development
I think it'd be really cool to hear your thoughts about Verse, the new open scripting language by epic. They seem to have very big plans for it as it's supposed to be the driving force for the metaverse, so entirely built with concurrency asynchronous functions and networking in mind
Always the Cherno Nailing it by explaning stuff for normal people like me, great job Man.
A small explanation.
This system is working based on base PCG class, which provides a lot of tools and ways to determine how to shape and drive logic within its guts. It is very similar to Blueprint geometry tools we know from 5.1.
@20:56 Those categories can be expanded by subclasses, so if you have a base class which contains this category "PCG", all variables you set to be in this category will display there as well. This way you can easily track specific groups of attributes.
Is it similar to Blender's Procedural Nodes ?
@@astroid-ws4py yes and no. No in a way that procedural nodes are just set of instructions, this blueprint logic can inherit from parent classes.
You can create PCG procedural rock formation, child class procedural rock formation with moss and then it's child class procedural rock formation with moss and plants. You can enrich the logic you have from parent class.
So glad I built a PC and left the consoles to the kids. Still think we’re years away from games looking like this but at least the tools are present
Nice to see a Game Dev react video! Love these. Great commentary.
I just love how unreal engine evolves with gaming epoch ❤
This looks like scam
I used to design/create race tracks back in the day.. We'd get the actual raceline playable and as close as complete, before building any of the other mesh - basically prototype it. It was the rise of modders + track editors that led to the more procedural development.
What tool/software do you use to be able to instantly draw on your screen like this ? Is seems amazingly useful
I think edge lets you do that
I've seen him do it in Visual Studio too, but now that gets me wondering whether it's a feature from some MS apps specifically
It's called "Zoom it"
@@matheusmarchetti628 Thank you !
I have also watched or read something about foliage rendering already was a part of unreal engine 5, so youre probably right about it coming out in 5.1
Support for foliage and glass-like materials was added in 5.1
The pace at which new features are being supported/added is impressive.
Maybe you could look at Counter Strike 2 and specifically the new and improved graphics? It's obvious that there are a lot of lighting improvements but it would be interesting to hear you break down exactly what we are seeing when we see "lighting improvements".
I've just realized watching this, I'm a game dev watching a game dev watching another game dev
21:54 . . . I agree and this is the tech I definitely would mind for any story driven game.
Is there anything you've seen that you'd look to take inspiration of for Hazel?
finally. i love the reactions
Yo did you see that Moonray, the Dreamworks rendering engine is now open source? Curious to hear your thoughts about it
Procedural generation uses assets that are hand crafted, no man’s sky works in the same fashion. The procedural is the way you code or programming tells the assets to work together to form your world.
Have you tried layering a decal material on top of glass? I'm trying to add a top decal layer made with substance but the problem is that the output node must be set to TranslucentGreyTrasmittance for glass to work. So it won't work... the decal is translucent. It would be good to know how did they manage to add those mud decals on top of the opal material.
7:30 I was in another tab and just heard this without context loll
8:57 The quality is so bad because of the CZcams compression algorithm, that screen was pretty “noise-like” which is very hard to compress, it has a high entropy so the quality for a given bit-rate drops
Yeah, anything black isn't utilizing Nanite. I think animated foliage is still not quite supported yet. I could be wrong though.
I'd like to know your opinion about writing our own engines, is still worth it ? I mean, a single person will never be able to get there
I think the black pixels in the nanite debug view is just a basic postprocess depth-difference outline shader just to make visualization easier, otherwise it's kind of just a multicolored soup of pixels
Or it’s due to the foliage close to the camera being GPU instances of the base asset, rather than nanite-simplified.
Typically when making a large scale physics engine, there are a interaction box that typically is longer in the direction that anything is moving. So if two boxes are intersected.
There is a close box for that is right tight to the true physics model.
And when the inner boxes touch they use the physics model. And typically artist designed. Some games just use a LOD mesh for it
Spheres are cheaper than boxes.
@@W1ngSMC That us true in some cases, The benefit of a box is that it can be streched in the direction of travle and also have a more exact intersection point.
The cost of the intersection points are much lower than the cost of the physics calculations
4:42 i feel so called out as i skipped forward and landed here lmao 🤣🤣🤣🤣
22:42 "yeah, so pretty beefy" what? isn't that literally the fastest gaming hardware in the world? lol
yes, but also like all other hardware its also way cheaper in 3-5 years when this tech is in finished AAA games.
you gotta keep in mind that these demos aren't optimized at all like a game would. and it's running in editor instead of a packaged build. Dev machines for this level of quality are always very beefy because by the time a project like this releases the hardware and software (like dlss 3) will have caught up
just a few years ago this would have had to run on an entire server farm way beyond the cost of a top tier gaming rig.
Yes, but for a workstation it's probably mid at best. They want to point out that their non-baked demo running in editor can run on consumer level hardware with that fidelity.
Wonder if the rt series is dead, last two episodes we’re just like a handful of lines of code
Great to see you back in a react video!
Nanite didn't originally work with foliage, but support for it is in demo stage (or something, I haven't been keeping close tabs on it).
I use to do testing for EA when they first started out selling online.
feels like we get a lot of tech demos, but never any actual new, graphically impressive games
They are literally selling an engine here, not a game.
@@vast634 I know, doesn't make my statement irrelevant
Very good ,So when do you think we are going to see games or even movies ect
This tech demo will probably be free when 5.2 gets an oficial release
"Why is this quality so bad" it's the encoding. It can't preserve a good visual fidelity for a given bitrate when the image is noisy. Nanite triangle visualization always does this, as will any kind of noise.
I'm pretty sure he knows this and was just being snarky, as someone who literally builds game engines, he obviously knows how images work.
@@unawakeful6931 That surprised me a lot tbh, but you can't know everything and it didn't make sense to make that comment willingly in the video.
It wasn't just a comment on that particular section, I think. The video was captured at 1080p, making the demo look blurry, instead of higher resolution, like 4K.
5 minute mark: Are people really complaining that you talk to much in these videos? Literally the reason I'm here is to hear your thoughts.
8:15 my computer scream due these amout of triangles lol
The truck is not 71 million, the scene is... and they do use polys as a shortcut that's weird...
today I learned about that screenshot youtube video extension
For continuous professional development points we call rolling through CZcams Attending A Seminar. If you leave a comment you attended a half day workshop.
I feel like they upped the brightness for the demo; everything just looks a little bit brighter than it should be.
I'm really eager to see how the UE Procedural toolset compares to Houdini Engine
10:00 there is something definitely weird with the wheel, cause in 5:19 that same wheel is much more round.
10:03 flat tire,10:06 back to normal state.
Which is weird,this engine is flaw.
youtubes quality goes down in the nanite view cause of the bad encoding youtube does, more detail (even worse when it's moving) destroys the quality
Isn't the concept of having to draw 'triangles' a bit limiting, especially if you can define a planar face with multiple vertices? It seems like the 'everything has to be broken down to a triangle' is a limitation of current GPU's ... wonder if anyone has considered the performance benefit of pure hardware rendering of planar polygons?
They have. They tried quads with earlier gpus the problem is that they are much slower to calculate than triangles.
@@blakkwaltz Thanks. That would explain why!
one point to consider is that when everything is a triangle the way to handle everything is the same (for parts of the process at least), which might lead to simplified special-purpose hardware as well as more easily parallelized works loads
Planes/Quads have to be divided into triangles at the gpu level because the triangle is the simplest enclosed 2d shape. Then you can make gpus to maximise efficiency on doing the same task at the simplest level multiple times at a fast rate aka rendering billions of triangles at the same time across the many gpu cores.
Its easier to do 1+1+1 many times with many students to calculate 1558693 x 27343083432 than making them do 358x6933 many times.
@@rudrasingh6354 like, I agree with you at the hardware level, but I'm pretty sure doing multiple "358x6933"s would be faster than essentially counting to 4x10^16, like, that multiplication plus a sum of the result should takes less than the time to count to two million or so : p
youtube compression on top of youtube compression...yeah i can count the pixels
2018 was 5 years ago,
Fu**, I need to get a grip on time man.
I actually prefer to hear your comments instead of just the video.
7:24 floating left rear wheel 🤔
They just say polygons so people can understand it better (marketing to n00bz), triangles might be too exotic for some :D
The black foliage is probably placed as static model by the artist, as a touch up.
The imported model will likely be done using quads, and Nanite directly processing those quads. Calling it polygons is more precise. The final triangles for rendering are just the result then.
@@vast634 The model itself can be created with quads of course, in the DCC tool of choice, but I do not think it is exported as quads.
The model's triangles are grouped by hierarchical clusters (using the METIS library), you dont need no quads for that.
Hey Cherno, can you review the Hellblade 2 new cinematic trailer as well ? It shows off Senua's facial expressions and what they've achieved with UE5 on Xbox series X.
So true, your having fun creating game systems and you remember you haven't even start making the assets that you're thinking of then the dilemma of Artstyle and rendering 😆
8:10 Whether the truck was 70mil tris or polys, he's likely referring to the source asset, and nanite makes it lower poly as needed, so it wasn't rendering many polys for that model as you can see large tris in nanite view.
What happened to the cherno 2 channel?
im 6 months new with unreal 5 , its RVT (runtime virtual texturing ) isnt it ? blending into the surface , they were vague as far as the procedural side , so we, ll have to wait for the blueprint for said procedural , & then swop out for our own asset elements inside blue print. is this right ? i downloaded unreal 5.2 peview , yesterday took half a day ! must be somthing new to view inside starter content !
i still fear the future where you pay 300$ for a game and 200$ is for the 1tb external ssd that is completely full with just one game. we won't have to buy dvds or so but external ssd/hdd to play one game.
I do think we go the other way. One reason games have grewn is banking. Also real world scans.
Pseudo real time lighting solve the first problem, and presedual map generation solved the rest.
On top of that back in the day (90s) games could be installed with diffrent levels of texture quality. Back then textures was 90-95% of installed game data so if you cut the texture from very high to high the install may go down from 600MB to about 200. Go down to medium and it was under 100.
I can se the same happen with nanite, while there is no global LOD there is a sort of LOD equivalent.
You don't want a 1.5TB install. Well go down from extreme to very high, and its 400GB.. Still to much, go down to high and its 100GB
While that is not a thing yet, I see nothing that would stop it.
@@matsv201 I'm not so much concerned about the textures themselves but the poli count per object. if they have millions of poligons for each single variation of small stones in the scenes and each object if needed or not has the same or more then textures won't be what drives up the storage space anymore. sure we don't need LOD models anymore but we are still stuck with the highest detailed model and we don't have the 1-3 million poligon per level barrier anymore before it slows down to spf and not fps.
i wonder how many Gb that shown scene would take after being compressed for storage and how many Gb Ram it takes to run it. 32 Gb of free ram probably won't be enough.
@@dovos8572 To my understanding it works like this.
Every object have a standard low poly wireframe model. Same as other game, that would be similar to a LOD-0 model.
On the LOD-0 there is a higher resolution displacement modifier with a 2 dimensional matrix table. I don´t know if its a 1, 2 or 3 channel table, my guess its that its 2 channel.
Normally a wire frame model need to be at least 16, possibly 32bit to have sufficient precision (professional system sometimes also use 64 bits).
But if the level under is corrected by the level on top quantification errors, that would be a problem if it was directly addressed. So Nanite can probobly get a way with 8 bit precision with no problem, actually i think it can get away with 4. (the same method is used in the original MPEG codec, so its really not something new).
This would be stored in a file with it segmented in different areas. So the deeper information is stored later in the asset.
Here is the kicker.. well there is actually two.
A matrix grid is to the computer identical to a texture. So the computer handle it as if it where a texture. Actually the vertex data can be mixed in with the texture data as a extra Chanel.
There is a few more tricks that can be made. Normally textures are stored RGB and sometimes with a few more channels like alfa or something.
But image data can also be stored as YC channel. This is how oldschool TV compresses color information. The reason why it have not been used in gaming prior is due to lack of mixed precision GPU. But now its pretty common, Nintendo Swich have it, and pretty much everything newer.
A HDR YC texture could be 16-8-4-4-4-4. Then you would need a additional 8-8 bit for the model matrix. That would give a pixel depth of 56 bit, while that might sound like a lot, that is lower than say a 16 bit HDR RBGI pixel of 64 bit.
So your 3 million poly object would be 3M*56bit... 21MB.. yes.. MB.. not GB.
The second kicker. Direct storage. Typically less than 1MB of the object would be in RAM. When you get close to it, the GPU read in the rest of the file from the SSD with direct storage. The thing is, you will never see both the backside and the front side at the same time, and really you would probably not even the side in full resolution. So the Direct storage would read in about 7 or 8MB of it when you are totally up close to it.
There is still one more kicker. Texture compression. With placing the bits a bit smart, you could fool the GPU that everything is a texture, so then you can use texture compression. Both lossless on the polygon position, and lossly could be used on pretty much all of the other channels.
(in theory you could use lossly compression on poligon-possisons as well, but then it get kind of complicated). That would shrink the object on file down to anything from a few MB up to a bit over 10.
And also, a stone or a pebel would not be 3 million polygons.
There is one more trick that can be used. Because every level of detailed is stored in order. You can pretty much cut the file at any level of detail and just leave it. It would still work.
Hey Cherno... if AI can now generate image based on their fragments, why not generate game worlds based on small fragments generated by artists or even text prompts.
If such a world could be generated in minutes, why not generate personalized worlds for players, based on what they would like to see.. perhaps this is the future of gaming
That sounds like an mindless ideas.You probally can create entire high grade game using ur.engine with its visual scripting for such a while already.The point is not about convenience,but versatility.Or more importantly,what makes us do art?When AI now can also create art to compete with us?Well,art is not about competition,it is in fact about emotion and love and meanings that we as human beings want to deliver to audiences.So let a machine,even a sophisticated one to do the entire work is less of ideal.
Epic is doing everything they can to eat Houdini's lunch.
have you seen the program Dreams that is a game Engine for PS4
14:50 Houdini?
Substrate doesn't work with path tracing yet.
You know what I didn't even notice that part you were talking about about how basically this generative level design is probably a precursor towards incorporating AI
because I mean unreal has all the quickso bridge and everything built in and if you had a chat GPT4 and somebody's already working on something experimental you could just have the AI organized and edit your assets your art assets you could basically make a game in no times flat because the truth is this you know most creatives just want to get down to creating you know eliminating all the issues of codeine and all this stuff it'll just make it easier for you to focus on and fine-tune your creative idea you know.
Wasn't nanite supposed to not work with deformation and animation? The whole truck is Nanite, going by that polysoup view, yet those wheels are deforming based on the pressure from the ground and gravel.
are they deforming? I only noticed them going up and down
@@user-sl6gn1ss8p iirc, they said they deform, will look for timestamp, but check out that part where they change the material of the car to that iridescent one and as the camera rotates, check out the front left tire that's rotated. It looks slightly compressed on its bottom side(9:59). Also check out when he tries to move, the same wheel gives the feeling its compressed as it turns forward, so the truck can move on.
Not sure what he says specifically, can't make it out, but in 7:03, he says "... simulates how the tires compress into form". He follows it up with Metasound but not sure if that's what he meant.
@@npatch oh, yeah, fair enough, I think the frames I stepped just didn't show the effect too much. Don't know much on nanite though, so hopefully someone can answer your question : )
@@user-sl6gn1ss8p Me neither, or at least not sure if sth changed since the first reveal, but I distinctly remember it was only used for static geometry, which is why characters, which both deform and animate, do not share that same treatment.
It is the same as AI generated images,impressive but you can guess,AI is so powerful now,it could even create real time animation from streams of video footages.
14:00 - Wondering if Promethean AI will be capable of doing that.
Really nice👍
This is exciting for maybe 7 years in the future. Right now i need to sell my house to get a 4090
Can't wait till we start seeing games taking advantage of ue5
What happened to your Java game development series? :( I feel like it needs an update :P
Wish you would watch it before or maybe wait until a segment is through to talk about it... several times you ramble on and on about something he literally explains right after you resume the video. Most of what you said is immediately worthless, if you would've just waited to comment on it after he demonstrated it.
hi, where i can download demo ?
this was great
Reactions 🙌
Can this run on my pc?
Can you do a reaction video on Counter Strike 2 and the Source 2 engine?
Likely that black foliage is skinned?
Will Hazel also have Nanite and Lumen?
Nanite and lumen are unreal engine specific.
That procedural editing stuff and all the megascans makes me worry that everything make with UE is going to all look the same - unless you put lots of work into tweaking it. It'll be like "oh, that looks like it was made in Unreal".
That would only happen, if everyone also just uses the same assets (which probably a LOT will do, because they just want use the hyperrealistic Quixel assets, instead of more stylized or low poly stuff). What this whole procedural generation reminds me, is Wave function collapse, or autotiling from some game editors for tilemaps. So while the underlying rules to create a world might always be very similar, the look will drastically change with the use if different assets.
There isn't any obligation to use any of these tools, most teams probably will use them more in backgrounds and the like at most and use actual custom assets in the more focused foreground, or maybe not use them at all so they can get a full style. While I definitely do agree a lot of developers just end up using these assets everywhere they can, I notice it's mostly newer/less experienced ones that just chose unreal engine because it's "the realistic next-gen engine", without regards for other aspects. And then in terms of many procedural things, as mentioned by the other person swapping for custom assets and setting up different procedural rules can have major effects on the style. But of course, there is almost always an underlying look in almost every game engine I have ever seen, regardless of assets used.
In some future when engines can create perfect realism, it all HAS to look the same, as it does so in the real world. (Talking about a scene in a real world biome / same Region). There is not artistic Interpretation in nature. Of course artistic touches will always be possible. Just as a nature scene in a movie will look different than one in a documentary.
I hope Nanite brings that much energy when running on a 4010 because the way Nvidia is pricing their GPUs, a 4010 is the only thing most people are going to be able to afford.
13:55 Yeah, github Co-pilot is already going to have voice control to guide your AI assistant. It won't be long before every tool has voice control. It is honestly insane that computers understand natural language.
Very cool
it's nice to have but nothing really new, studios build such procedual generations since years in their own engines or even in Unreal itself.
I hope 5.2 offers more
Should unreal engine divide video game industries? if people saying unreal engine photorealistic games ruins video games. The truth is I don't know what to say. I support Unreal Engine developers, photorealistic artists, photorealistic game artists, photorealistic game developers, etc. Is it easy to make photorealistic game? yes, quixel, photo 3d modeling, photo realistic assets saves people's time. I don't know why some people are against photorealistic game. I'm not against it. I see bob ross's painting. but beautiful and realistic.
Not too keen on their decision to include an SUV sponsorship, but the rest looks cool.
It's a car bro. Relax.
@@marktime7692 It's not the individual car that's the problem, it's the ongoing and unnecessary promotion of inefficient, dangerous vehicles (SUVs, I mean) during a climate crisis. I'd just rather they found a different industry to sponsor.
11:00 everything is for CGI. Nothing for games actually. They even did not mention games.
So people should calm down.
Loads of tech, but the games implementing all this are not out there.
Outstanding work by the artists and your statement that the engine won't be worth s^^^ without great assets is indisputable. But when AI companies that create art generators gets their hands on these models....what will happen to the industry, most importantly, the people in it?
I really think AI and procedural generation is something that should be frowned upon and I agree with Valve for removing games that have AI art in them. Games are supposed to be worlds that are created completely out of the artist/game designers minds. Using AI will quickly destroy the character and soul of games. If that is truly the future then we should really cherish the media and games that we have now.
AI is a blessing for indie/solo devs, nothing to be scared of, it's just a tool for artists/devs/programmers, if your entire product is made by AI ofc it will be soulless and generic.
Man, one serious downside of these reaction videos is when you're like "yeah idk, he didn't really explain it" only to then continue on with the video where he explains it. Is there any real reason to do a "first watch" video rather than watching it beforehand and being able to give a more well thought out commentary on it? It's not like you're having any crazy reactions to seeing things for the first time that couldn't be replicated on a second watchthrough.
I can understand the desire for realism, but I game to get away from realism. I want to feel like I am in a different world.
be fun to see your reaction / thoughts on the new Zelda gameplay / Fuse mechanic