Why are most PBR Textures only 2x2m ? Can someone enlighten me?

Hi guys,
maybe someone can help me understand this:

Sites like Quixel and Poliigon have amazing PBR materials. But almost all of them only have a 2x2m real world size.
This is great for smaller objects and close up’s. But for bigger objects like facades, walls, roofs, grounds, roads and so on it means, that they have to be repeated/tiled a lot. And every tiled material needs at least one overlay (dirt, surface imperfections, leakage) to make it look decent. Plus some untiling technique to get rid of the pattern.
Even in the wider Quixel demo view (the one on the right), the tiling gets very obvious instantly.

Why don’t these sites additionally offer textures/materials of 4x4m or 10x10m sizes? (or even 6x20 for roads).
This is something I don’t understand. Especially for ground, plaster and road textures.
Not everybody does close shots all the time.

To be honest, this is the reason I did not subscribe to Quixel yet. Even though I absolutely love their stuff.
Instead I shoot a lot of textures by myself. In much bigger scales. Close to what I need them for.
For example, it would be extremely difficult to get a street texture to look like this, if you start with a 2x2m material:


This is a reduced version of a street texture I shot. It’s about 20m long. The original files are 8K (with all the norm, rough, disp maps.) The texture is tillable and has a feathered alpha at the far ends, to use with arrays or instances.

Why don’t the big sites offer materials in such sizes?
It would be so much easier to create realistic looking wider shots, with such textures.
Is there some reason I don’t understand?

Please don’t get me wrong, I don’t say they should make all textures bigger. I only ask for additional, bigger texture sizes. And I don’t mean more than 4K or 8K. I only talk about the actual, real world size of the textures.

I would really appreciate if someone could explain this to me.
Am I missing something?

Answering from a videogame point of view…It’s purely about being practical and optimal.

It is not difficult to get the same result with 2ks, all games do so by combining maps together in-engine.
Can be through use of decals, through vertex painting to blend your materials etc.

This way you have way more content, can cover way more kind of areas etc.
There is no universal rule do not get me or what you find online wrong. You can make your scenes however you want… But the reason why everything is always divided, packed, 2k… Is reusability, flexibility, optimizations etc.

If I had to make the ground in your reference photo ingame, I’d probably do:

  • 1x concrete materials (with the seams) + vertex blendable with a dirty version of itself + different grungy blend maps maybe.
  • 1x decal sheet covering all ground details (yellow stripes, square cuts, holes, black patches, some seams so I can modulate the floor a bit however I want here and there.

So yeah. Just break things down sometimes… it will help you go faster in the future.

1 Like

First of all - thanks a lot for your answer Anothercaveman.

That’s exactly what I mean. You need several textures, maps & decals for the same result. All in all this takes much more time, than getting a finished texture, like the one I’ve shown, from a service like Quixel.
I can imagine that your way makes more sense for videogames, as you could use the concrete material and grunce maps for other things as well. And can easily create different version of it.
I can understand that when starting with a concrete material.
But what if you start with a texture which has very obvious structures or details in it. Like wallpaper or paint peeling off a wall? They quickly get very obvious if you tile them. Even adding dirt will not help that much, as the structure of the texture is so obvious and strong already?

I do sfx for feature films and create photorealistic sets - and here ‘big’ textures speed up the process dramatically. They simply offer more variety and ‘life’ - and finally create more realistic looking results - quickly.
The more ‘reality’ I use for CG - the easier it is (at least for me) to make it look real. ‘10m of reality’ is different than ‘5 x 2m’ of reality. :wink:

But I totally see your point as well.

Agreed on some points :slight_smile:
To avoid the repetitions well we cheat all the time aha. But vertex blending and decals can do a pretty good job. Especially when driven with blend maps etc. Good example here https://www.artstation.com/artwork/x2zRO

Now you can also do all that to in the end bake a super wide texture.?
I get you and how nice it is to be able to drag and drop huge areas directly. Kind of feels like photobashing in 3D aha

Did you check out other alternatives to megascan maybe? Not sure I’ve ever seen what you’re looking for in wide quantity tho. good luck!

I’ve bein cruising around on loads of texture sites. But it’s more or less all the same. It’s no big issue. I simply wondered why the texture sizes are always so limited. Your explanation for game purposes makes perfect sense. And I think the gaming industry is a major client of sites like Megascan. Oh well, I guess I’ll continue to shoot my own textures…that’s cool as well. :movie_camera:

My simple take(also from game art side of things, mind): it doesn’t matter whether you are dealing with 2k or 8k maps, you are still dealing with finite resource, and at some point will have to go procedural, effectively getting practically infinite resolution and a lot more variety. While also keeping your source material on storage media down.

Hi Felix, just to avoid misunderstandings - I don’t talk about 2K or 8K. I mean the real world scale of textures. 2x2m or 4x4 meters for example. Obviously, the smaller the real world scale - the more often you need to repeat/tile the texture.

I can’t judge it from a game design point of view. Of course I trust you there.
But I simply have never seen any procedural texture which looks perfectly real, organic and authentic.

For lot’s of things it’s so much easier to work with ‘photo’ textures.
eg. On this picture below, the storefront blinds (don’t know the exact word in English) are just simple, flat planes. But they look good, because they have real, photographed textures on it. No need to model much, no need to tile a texture and add overlays/decals. Just add this one photo-texture. Simple and easy. There are so many little details and imperfections in reality, which need a lot of work to recreate artificially.
Of course this is a special case, and I mention it, cause it’s an obvious example.
But it can work similarly for other things.

I totally understand that you use procedural textures for games, to keep the data sizes low, though.
And that’s perfectly fine.
What I don’t understand is, that sites like Megascan, which have thousands of textures, don’t offer at least some versions of their textures in bigger real world sizes than 2x2m, for big ‘objects’ like facades, grounds, floors, streets.

But hey…it seems like I’m the only one who thinks like that…so it might be probably my issue. :wink:

Also in movie industry, it is common to just shoot using camera in real like locations or sets and use CGI where you need put some spaceship or something. Also it is common in movie CGI to scan real environment (or even miniature model) to get assets.

Those texture collections are kind of basics and intended to use as tools to craft bigger environments.

When you do rendering, it doesn’t matter is it game or movie, all stuff needed to fit memory that can be accessed fast.

I think there’s one other big reason to use tiled textures with variations and decals: art direction. It can potentially be much easier to art direct a material made up of modular parts, for example a couple variations on a tiled background, plus manually placeable and easily movable decals for variation.
If you use a single giant texture for things, especially for items that aren’t simple planes or other shapes where it’s easy to pull into photoshop and understand what goes where, iteration time for material variation can go way up when tweaking, even if the initial implementation is faster.

That may just by me, though.

Actually I am in the movie industry. :wink:
And there are lot’s of set extensions used. Lot’s of locations you would think are real are only CG. To build sets in CG is much cheaper than to build them for real. Sometimes it’s even cheaper to create locations in CG than to have a whole crew travel to a location.

Well, I think it matters a lot. For games you need to be able to render in realtime. Also on mid-range or even low-range PC’s. (Unless it’s a prerendered cut-scene.) For movies you can take much longer to render. And you can use workstations. The quality needs also to be higher for movies.

Well in terms of ‘real world scale’ you would likely have different maps, the masks you might use to mix different smaller scale detail layers would obviously be larger. The exact scale you would use would vary depending on the nature of the element you are working on/creating.

In as far as the Megascans it’s probably more to do with their pipeline for creating the assets and the cost involved. It’s probably cheaper to have a strict target scale to work with. Less issues and surprises that come up.

Yeah, I know that there is CGI sets used from 90’.

Actually this may be even possible to specialize to just create accurate locations to other industries. Like how many movies and series happens in ancient Rome or some New Zealand environment? It should be possible to download as huge blend file and detailed scenes from some buildings.

Sure movies can have a lot more rendertime and easily more complex composition.

But the thing is, that actual rendering need to be done where CPU/GPU access some memory with complex datastructure so the real bottleneck is the size of that memory. If everything in the renderpass or partial rendering doesn’t fit on GPU memory, you need to render using CPU. If everything in renderpass doesn’t fit in RAM, you need to swap from harddrive. When assets doesn’t fit in memory, there is huge slowdown.

Yeah, there may be some 2mx2mx2m scanning device made what they carry around, and automated process that when they press button, they got asset files.

And if they scan something bigger, it is some special case that need manual labour and some different scanning method.

That’s a very logical explanation! Makes perfect sense. Haven’t thought of that. Thanks.