Texture 2048-->1024

[leakingpaint]

Howzit everybody

Any tips here, I started creating textures for some custom models at 2048x2048 and downscaled to 1024x1024 the results are less than great

Please check the pics below to see what I mean, the walls are where I'm concerned - any help will be greatly appreciated.

Pic01_1024
Pic01_2048

Pic02_1024
Pic02_2048

thanks

[chorlion40]

i'm gonna guess you used photoshop to downscale the tex?
or perhaps a function in the editor?
well don't because that is what is causing the blur.
use MS Paint :P
it should get rid of the blur
*peace out*

[nightFlarer]

Don't know if this is your problem but if your saving the image file after downscaling don't save it as jpeg or something like that, because the compression makes it look bad, save it as tga.

[leakingpaint]

well don't because that is what is causing the blur.
use MS Paint :P
it should get rid of the blur

Ok I'm officially confused here, Why would paint do this better than photoshop? I use Image Size-->Bicubic Sharper (best for reduction) Then I apply a slight Unsharp mask and save it as a 24bit TGA.

As far as I know paint doesn't support TGA and opening a 2k image in paint should prove interesting...

[.newzero]

Ok I'm officially confused here, Why would paint do this better than photoshop? I use Image Size-->Bicubic Sharper (best for reduction) Then I apply a slight Unsharp mask and save it as a 24bit TGA.

As far as I know paint doesn't support TGA and opening a 2k image in paint should prove interesting...

you still try to reduce the texture size which causes this blurred mess. one thing you could try is to increase the texture detail settings in ut3. also you should give detail textures a try which can visually improve the material details.

[bazzwano]

Ok I'm officially confused here, Why would paint do this better than photoshop? I use Image Size-->Bicubic Sharper (best for reduction) Then I apply a slight Unsharp mask and save it as a 24bit TGA.

As far as I know paint doesn't support TGA and opening a 2k image in paint should prove interesting...

with photoshop I always have to sharpen the images after I re size them, I know it sounds crazy, but MS paint does actually size an image very nicely.. it remains very clear and sharp.. then take the scailed image into photoshop..

now Im sure that your problems are more then just how your sizing the image. But ill leave that to the experts. Im only just beginning to add textures into UnrealEd

[FakeTruth]

I see that on your texture the ONLY extra detail is some grain. Instead of using crazy high resolution textures you're better off with importing a normal map for a grain texture, scaling it down through the material editor (for even higher resolution) and adding it to the existing normal map, or if you didn't have a normalmap you put solely in there. This'll add to the detail without increasing the hardware requirements for your map too much

[SynaGl0w]

For me, the editor and PIE always displays textures at half their normal resolution regardless of LOD group settings. 1024 displays as 512, etc.

[Oddside]

In order to resize (shrink) an image without any blurring you need to use the 'nearest neighbour' algorithm if available. I'd hazard a guess that MS Paint uses this algorithm. The only downside is that this algorithm can lead to moiré artefacts in your textures.

[leakingpaint]

Thanks everybody for the advice. Surprisingly paint does wield some pretty good results, lol.

"nearest neighbour algorithm" does perform this function as well, thanks again

Lol, paint...classic...

[bazzwano]

nearest neighbour algorithm
where did you find this option in photoshop? I normally transform but that doesnt look good. and when I scail, it crops the image.

[leakingpaint]

bazzwano the nearest neighbour algorithm option is available if you click
Image-->Image Size. At the bottom of this dialogue is a drop down list which enables you to choose an algorithm before resizing but I'm afraid it doesn't help in my case see below.

Edit:
bicubic sharper / nearest neighbour / paint or photoshop make no real difference when you downsample the image. I thought it did but even after doing a full rebuild and saving the level and package if i reopen the editor the textures look shoddy yet again.

I thought it was fixed but it only appears to be until you close and reopen the editor.

If you double click your imported 2048 texture within your package and then go to LOD Bias you can type 1 which, should, convert the texture to 1024 when you publish the map.

Setting the LOD Bias to -1 looks great in the editor but I think it's a bad idea performance wise. I'd really like some clarity on this if anyone can help.

Thanks

[DGUnreal]

Regarding resampling algorithms, PhotoShop depending on version will support a number of them. Upsampling and downsampling often require the use of different algorithms to get the best result.
Nearest-neighbor simply drops pixels when it is downsampling, so a 50% resample only retains every second pixel. This can work ok for certain types of textures to retain sharpness, but you completely lose half of the texture information.
Other methods from bicubic to lanczos use differing types of pre-filtering, which reduce aliasing in the image but tend to blur it. Pre-filtered algorithms also usually take a weighted downsample result of pixel groups, so you don't lose much of the original texture information like you do with NN. These algorithms should be used on images with high-contrast edges or any image with alpha channel data (NN really messes up alpha).
Wikipedia has some good info on resampling, as do some of the Photoshop tutorial websites.

Don't go by what you see for textures in the editor when you save, close and re-open. The editor will often only stream in a lower mip until you click on the texture in the generic browser. You have to play the map and see if it looks ok there.
The editor's settings for TEXTUREGROUP are also different than those in-game. If you look in the My Games\...\UTGame\Config\UTEngine.ini and find the [SystemSettings] and [SystemSettingsEditor] sections and compare LODGroups you will see that smaller texture mips are usually loaded in the editor compared to the game. This is to reduce the editor memory usage but still provide the proper texture resolution in-game.

What TEXTUREGROUP did you use?
If you used World, which is the default for world geometry, then it has a MaxLODSize of 1024, so even if you imported a 2048 texture, it will still only render at 1024. This is regardless of whether you specify a LODBias of -x in the Texture Properties since LODBias is calculated before being clamped within the Min/Max LODSize.
If you have to use a 2048 texture you must place it into a different LODGroup that supports a larger MaxLODSize such as Skybox. Be aware though that the LODGroup you choose can affect the lowest rendered mip.

Now concerning the fact that a 1024 texture may look blurry in-game also has to do with the UV mapping that you are using. If the texture is scaled so that it is large compared to the world and therefore large in respect to the camera frustum height, each texel will render as a large square when you are up close to it. But go by the in-game look, not the editor look, since the editor LODGroup may be rendering a smaller texture mip.
The best way to fix this if you can't change the UV scale, so long as it isn't massively blocky (like an old Quake game), is to use a Detail NormalMap in your Material. You can find info on these in my UDN articles on Epic's site. There are a bunch of NormalMap Detail textures available in two game packages. If it is massively blocky, then you either have to use a smaller UV or higher resolution texture, but in most cases a 1024 should be fine.

Hope this helps...

[leakingpaint]

Thanks very much for this interesting information.

I think I understand for the most part except for the UV scale? When you say UV scale do you mean the actual rendered UV within the 3d program? The size of the UVs within the zero to one area? i.e. the largest parts of the mesh need the largest area within the UV Space?

How would one determine if the texture is large compared to the world camera?
Would using a 28 / 35mm camera to preview the texture on the mesh within the 3D program be better than say using the perspective viewport, or is it simply trial and error until you find a texture which fits the world and the mesh well enough?

Here are some progress images in case anybody is interested
http://i51.photobucket.com/albums/f3...t/tex_mush.jpg
http://i51.photobucket.com/albums/f3...x_mushwall.jpg
http://i51.photobucket.com/albums/f3...mushwall02.jpg

Thanks, forgive the questions if they seem dumb.

[DGUnreal]

I think I understand for the most part except for the UV scale? When you say UV scale do you mean the actual rendered UV within the 3d program?

The UV scale applied to the texture coordinates on the mesh in the 3D application.
Depending on how you UV'ed the mesh, and whether the texture looked correct in the 3D software, you may be seeing differences once it is imported and used in the map. Also if you increased the StaticMesh.DrawScale then that also increases the size of each texel, making the texture look blockier as the mesh is scaled larger.

Take a look at the stock trees in ASC_Foliage.SM.Mesh (Tree02, 02b, 02c, etc.) and zoom in fairly close to their trunk and branches and you can see how low-res and blocky that texture really is.
In order to keep the wood grain at the proper scale though, the only way to make them less-blocky would be to use a higher resolution texture. Changing the original UV scale in this case would only make the wood grain too small.

Or even better, look at the building city blocks in the HU_City package. They are really low-resolution textures.

I assume that with your mesh, you can get right up to it in game. In this case, you have to decide whether you can change the original UV in the 3D app (you may not be able to depending on the source 2D texture used since scaling wood grains or rock textures etc. too small will make them look funny), whether you want to use the extra texture memory for a 2048 texture, or whether you just want to fix it up a bit with a Detail NormalMap in the Material.

You can use the Texture Density view mode in the Editor to see how one mesh's UV size compares to the other objects in the scene. The meshes with colors moving towards orange-to-red are using a smaller UV size, so if you get in real close in the editor you'll see in Unlit view that a mesh with blue Density has a much larger blockier texture than a mesh with red Density.
This view mode is to allow you to balance out the texture scale across the map so that things look consistent.

How would one determine if the texture is large compared to the world camera?

If you can walk up to the mesh in-game and it looks really blocky, then chances are you need to use a smaller UV scale if possible, or at minimum apply a Detail NormalMap to try to hide the low-texel blocky look.

... or is it simply trial and error until you find a texture which fits the world and the mesh well enough?

Usually trial-and-error until you get used to a good reference UV scale for meshes. If you are using something like Max and have the world scale set correctly so that it matches Unreal, then meshes should look almost identical between the two.

The mesh location on the map and in the scene can also determine the texture resolution and its UV scale. If the mesh is a tall building way in the distance that the player can never get to, there is no reason to use a high-resolution 2048+ texture that looks crisp when you are standing next to it -- cuz no one ever will be there in the game, so a low resolution texture can be used.

The issue that you will always run into is that you often have to settle for slightly lower texture resolution in games because of limits on texture memory.


Probably more info than you wanted... and probably didn't directly answer your question.

[leakingpaint]

This information is very useful particularly the Texture Density view within the editor. I found this most interesting.

Probably more info than you wanted... and probably didn't directly answer your question.

The more information the better. I hope to do this for a living one day so informative feedback regarding anything is greatly appreciated.

I'm going to try play with the UV scale within 3dsmax and try use the Texture Density view to better match the detail. I hope to refine my skills, thanks very much.

[chorlion40]

i see walls of text but all we needed was one word. paint.
see how it works?

[DGUnreal]

i see walls of text but all we needed was one word. paint.
see how it works?

Perhaps if you had read the "walls of text" you would have known that isn't the issue.

[leakingpaint]

While I thank you for your input chorlion40, I requested this information from DGUnreal and am thrilled at the amount of detail he has covered regarding texturing. It interests me to know all aspects of whatever I'm currently busy with.

Also the issue in my case was a combination of what DGUnreal pointed out (UV Scale / texture density and the diffuse I was using simply had too many fine details)

Paint does not re-sample images better than Photoshop I was simply using the incorrect algorithm which in this case, as Oddside pointed out, should be 'nearest neighbour'

Thanks again for your suggestions and help.

Updated screens - excuse the lighting this is simply to test what the textures look like whilst being lit

http://i51.photobucket.com/albums/f3...int/tex_01.jpg
http://i51.photobucket.com/albums/f3...int/tex_02.jpg
http://i51.photobucket.com/albums/f3...int/tex_03.jpg
http://i51.photobucket.com/albums/f3...int/tex_04.jpg
http://i51.photobucket.com/albums/f3...int/tex_05.jpg
http://i51.photobucket.com/albums/f3...int/tex_06.jpg

[bazzwano]

Thought It might be worth showing that you can make good textures even with a 512 res texture, so long as you have good normals. Im still working on this, I have been painting it from scratch (no textures), but this is a 512 res texture. thought I would show that off



Good info as usual thanks DG.

I have a question. If I were to transform an Image to fit the canvis, how do you set these resizing options because you dont have any when you transform an image.

I know thats these options have nothing to do with your issue, but I would like to know how to transform a nice sharp crisp image in PS without the bluring. SO I have learned how to scail the image but what about transforming the image by dragging the corners.

[leakingpaint]

In PS if you click Edit-->Preferences-->General you should see Image Interpolation options which I think control how an image is resized when using the transform tool. I'm sure you already are, but if not, make sure the document you're painting your texture in is 300dpi which helps for resizing when using the transform tool. Even if the document is 2048x2048 I find working with 300dpi better...although it does bog my slow ass computer down.

It depends on what you're making a material for. A tiled texture for say a wall or floor can quite often be very low res but as soon as you start texturing an object, say for example a vehicle or a weapon placing all your detail into one texture, makes a 512 res image a little insufficient, not that it wouldn't work.

Another alternative which I think would work quite well would be to blend two materials together especially for things like moss / dirt etc.

[bazzwano]

Hey LeakingPaint

Im wondering how you got on with this. I was reading this on eat 3d. I see you posted some pics where it does look allot better? It looks like the material was just to large for the size of the mesh yes? I dont touch these LOD settings, but what did you do... Just encase I have this issue for my large meshes.

Thanks

[leakingpaint]

In my case, my UV layout was too large. For example I was designing a very large mesh but keeping the polygon count very low which caused the UV layout to make my texture look a little blurry.

When creating meshes like I was creating the solution was to increase the number of polygons so when I unwrapped my model the texture I applied wasn't as "stretched"

This isn't the best solution as you have to increase the number of polygons but it did improve the look of my texture a little.
I think using a model with slightly more polygons and a 1k texture
is better than
using a model with fewer polygons but a 2k texture.

It really is trial and error until you get something you like. I now make a base tile texture first and apply it to my model and check it looks alright before I waste too much time painting something that doesn't work.

I also found that making 2 specular maps worked well. I created a specular map to match my smaller detail normal map. It's a combination of patience and persistence which helps as I'm sure you already know.

I wish you the best of luck.