I was working today on a scene and noticed that my Firefly renderer started doing the usual lock up when rendering. All I had was a bedroom, 1 vicky figure, some furniture and some clothes. I thought, "nah, couldn't be that intense" but looking at my poser 5 memory usage I was a 1.25 gigs! Yeoch. So I go through the textures for this scene, which I got from purchased packages and here's what I find. 1. Textures that describe the rail on the bed and the feet, mostly all white with just a little art in the upper part of the texture:2000x1380 YEOCH! 2. All textures using repeating tiles: 800x600 YEOCH! 3. A simple reflection map used for a lamp: 2500x3000 YYYEEEEOOOUUUUCCCHHHH!!! So I go through and reduce all textures to reasonable sizes and powers of 2 and voila, my scene renders just fine. So I want to post this to once more remind texture makers to be careful with their texture packs and sets. Let's go over a little math to help people understand how this can get crazy real quick. Texture size goes up exponentially, not linearly. That means doubling a texture does not double its memory space, IT GOES UP BY 4 TIMES! Here's a break down of texture sizes assuming Poser uses 32 bit (4 bytes per pixel) texture memory. 128x128 = 65,536 bytes 256x256 = 262,144 bytes 512x512 = 1,048,576 bytes 1024x1024 = 4,194,304 bytes 2048x2048 = 16,777,216 bytes 4096x4096 = 67,108,864 bytes and the mega 8192x8192 = 268,435,456 Now the assumption here (and based on my knowledge of rednering engines and graphic systems) is poser doesn't give a hoot you have that whoppong 4096x4096 all crunched down into a 128k jpg file. Poser will still expand that texture out to a block of 67 megs of ram to rasterize the texture (work across the texture to get the bits so it can display the texture on the polygon). So that means a 4096x4096 texture of solid black takes up as much room as a texture with a picture of France on it. Of course, I'm assuming they unpack textures in Firefly which is a pretty reasonable assumption. So what does this mean for texture makers? Well having a skin pack with a 4096x4096 skin texture and adding lipstick, eyeshadow etc packs at that same res. will blow out some poor saps poser memory worse than squeezing an elephant into a phone booth. Making the texture for wheels on a bed be 2000x2000 pretty much dooms that bed to a lock up once it tries to render. Lets add to that most rendering systems work in powers of 2 for texture sheet sizes. They do this since the math to rasterize the texture is far faster when the texture is a power of 2. So what does Poser do with a 1100x1100 texture? That's not a power of 2 (2,4,8,16,32,64,128,256,512,1024,2048 etc). Well if Fire Fly is like any other rendering engine it creates the next biggest sized texture sheet that IS a power of 2 and puts that 1100x1100 texture in the upper left corner and rasterizes that. That means that 1100x1100 texture turns into a 2048x2048 texure! YEEEOOOUUCH! All that wasted space, worse if that's just the wheels on a bed or lipstick on a person. I have no idea if poser fire fly has a sheet management system which might use the unused portions of that 2048x2048 sheet to put smaller textures on it but at the rate most texture makers are going, there AREN'T ANY small textures to go there! So not only is size important but keeping the size to a power of 2 is important. Now lets talk about another issue which is very important that appears to be missed namely texture resolution. If I render an 800x600 picture with a flat billboard with an 800x600 texture on it, unless the camera is set up in such a way as to see the whole billboard, I'm not going to do any better than 800x600 texture resolution. If I zoomed in on that billboard so the full screen is encompassed by half the billboard then having a higher res texture would give the render more resolution. But for the most part people don't render zoomed up on an object. They render several objects in a scene. So what does this mean? It means that having a 4096x4096 face texture does you no good if your final render is 800x600, especially if the face only encompasses 1/4 or less of the total area. The computer can still only render one pixel on a pixel spot. It can't subdivide pixels. So You could use an 800x600 (or a 1024x1024) skin texture in its place and see no real difference. Imagine a crowd scene rendered at 800x600. You see 12 people in the crowd. Now, if each person's face texture was 4096x4096, you wouldn't see anymore detail than if that face texture was at 1024x1024 or even 512x512. In other words a high res texture does you no good if the item rendered comes out less in size that the texture resolution. So to sum things up, here are some helpful tips in texture making that will guarentee less P5 fire fly crashes and lockups. 1. Keep textures small! I know you're proud of a texture but a 2000x2000 texture for a bed wheel is absurd. I have a standing rule no textuire can be over 1024x1024. 512x512 is the standard and most are in the 256x256 range. 2. High res textures do you no good when the item rendered is less than the resolution of the texture. A face rendered on a 800x600 sized picture won't benefit much, if at all, from a 4096x4096 skin texture. 3. Keep textures at powers of 2. (2,4,8,16,32 etc). Poser is more than likely stuffing your 1100x1100 texture on the next size up sheet that is a power of 2. That's a huge memory waste. Also try and keep textures rectangular. Some systems don't like non-square textures. 4. Disk usage is irrelevant. A 50k jpg of a solid black 4096x4096 texture will still take up 67 megs in ram. YEEOOOUCH! 5. If you're selling to P5 only users, try taking advantage of the P5 shader node abilities. That means insted of making a brick wall with a huge texture on it to get a detailed repeating wall texture, you can set up the shader node to repeat a much smaller wall texture for you. Also using procedurals and blender nodes can greatly reduce texture usage.

I may be misunderstanding it, but doesn't the "max texture size" in the P5 render options limit how large that "sheet size" is? So that, for instance, if you have a 4096 X 4096 texture and your "max texture size" is 1024, Poser will resize the texture to 1024X1024 and only use 4MB instead of 67?

FyreSpiryt, you are correct. These are all outstanding points. The drive for "photographic quality" is behind much of the texture bloat. The overwhleming force in action is still Poser 4, as well, which actually works better with higher resolution textures becuase the rendering engine isn't as advanced as firefly. From what I've read, the standard texture map size used in Pixar's work is 1024 at the largest. Most of them are 512 or so, and the shader system is the key element in using them. For my part, I use about a 2048 size texture at the largest, and that's for the focal point. One issue to keep in mind, as well, is that the templates for much of this stuff are rarely created in powers of two -- the V3 templates are some of my favorite examples of this. Totally off the wall there. ya do what you can. That bed texture seems particularly interesting...

Excellent points - and yes, FyreSpirit, the "max texture size" in Render Options does scale the texture during the load process. I keep mine set to 1024x1024 most of the time, and I have very few problems rendering fairly complex scenes with lots of large textures. This DOES increase the load time, however. I think the high resolution trend in textures to be bordering on the absurd, especially by those vendors who just fill the sucker with a few solid colors and little detail here and there. They might as well have a 256x256 texture for all the detail they have - it is just a monumental waste.

And of course, if you're touting a 4000x4000 texture map, you're going to be equally proud of the 4000x4000 bump map that goes with it ;) Extremely useful info Mason - I had no idea of the importance of keeping texture sizes to powers of 2. I think I feel a bout of texture resizing coming on! - gps

I didn't know all that (of course since I'm still new to 3dWorld) but I have a question. One of my friends renders parts of a scene at a very large size because then she brings all the parts to PSP to create a poster size (24"x36") image. For those who do what she does, if all the texture makers would create their textures at 1024x1024 or less, it would be a problem, wouldn't it? France

Francemi , yes, but only to a certain extent. IT's going to depend not on the final size of the composite, but ont he size of the actual render window itself. IF she can render at 3000 pixels square on her system, then yeah, a 1024 is gonna hurt. I've routinely rendered images 2000 pixels square at 300 dpi, and there is a bit of the ickies when the main textures aren't large enough (roughly the same size as my rendered image). And by those, I mean the ones that areon the objects in the focus of the image (for most poser users, that's typically gonna be a person, lol). There is a reduction in the "reality", so to speak. In poser 4, it's pretty much a lost cause. In P5, though, I can use shaders (and lights -- lights are all important) to return a great deal of that reality. And of course, I can always use the original

I really don't mind the large dimensions of figure (head and body) textures - the resolution can be handy for portrait work and I can resize them down (and I do most of the time), so I wouldn't want those vendors to get the wrong idea. The thing that bugs me is what appears to be a trend in props and accessories toward large texture sizes that are created by SCALING UP an otherwise small texture - apparently just so they can advertise the larger texture size. Unfortunately all these textures do is waste resources. So to those who make textures, if you have to size-up (enlarge) a material just to make a higher resolution final MAT, they you are just wasting everybody's time and resources.

Here here, ronstuff! I have some eye textures on my drive that are HORRIBLY pixellated. One I just popped open. It's 1500 X 1500, but it's made up of BLOCKS that are 6 pixels by 6 pixels. It looks EXACTLY like a 250 X 250 texture that's zoomed in 600%. Why would you even do that? (Well, you just said why they do that, but it's stupid.) While I've got it open, I'm gonna go resize that thing.

People should make their own textures. That way they can be any size they desire.

Honestly, I think that it is better to have large textures. When I download an item, I check what is included with it. If the texture is too large or if the file size is too much, I can resize it or save it as a jpg of less quality. If the texture is too small or the resolution too poor, there is nothing I can do. France

Thanks for the info! I make lower resolutions version of the same texture sometimes,just resave it with a _LR at the end,in the same directory, open the MAT files witn wordpad replace the names and I have a MAT file of the LowRes version too... And I always appreciate when trans maps are in the texture, however the best option is transparent part to have their own map like V3... pitklad

Hello Mason, I have been having similar render problems with FF and have been reading through these threads. Would you please clarify something? You talk of reducing the size of texture maps. By doing so, will they fit an object properly? For example: a body texture map that is 2k by 2k reduced in size by half, 1k by 1k, will everything line up properly on the corresponding figure? And, does quality suffer? Thanks, Larry-L

I would rather have high-res textures and reduce them myself then have merchants sell 1024x1024 textures. Lo-res textures are fine for images shown on the web but they're not good for when you're doing print work.

Larry-L: UV mapping of the object has nothing to do with the dimensions or aspect ratio of the texture map. The UV coordinates are values from 0 to 1 on the U and V axes. The pixel in the lower left corner being 0,0 and one in the upper right being 1,1 - all points in between are decimal numbers like 0.02376981, 0.52096723. Through simple mathematics, a 3D rendering program will take ANY texture map of any dimensions or aspect ratio and apply the lower left pixel to UV coordinate 0,0 and the upper right pixel to UV coordinate 1,1 and all in-between values to their corresponding equivalent. So it does not matter whether your map is 512x512 or 2048x2048 or 300x900 or 3456x7654 or whatever, it will be UV mapped according to the relationships established in the UVs of the geometry, and not by the texture coordinates. In other words, UV coordinates are based on PERCENTAGES and not on pixel points - this is why they are called U and V rather than X and Y. So UV coordinates represent the location by describing what percent (expressed as decimal values) a point is between 0,0 and 1,1. It is really easy to visualize if you think about describing a point on an image map that is 15% up from the bottom and 25% across from the left. The dimensions of the map itself would not matter because the percentages would always point to the same location of the map regardless of its dimensions or aspect ratio. Jade_nyc: I think the point being made here is about things like V3 eye textures that are being distributed at 1024x1024, but they are nothing more than a 256x256 map that has been scaled up. That is just a waste. And who needs a 1024x1024 eye texture anyway if the head texture is 4000x4000 then a 256x256 eye map would be about the same scale (talking about 3rd Gen Mil figures with separate eye maps not 2nd gen where the eyes are on the head map - boy what a WASTE that is to have separate eye textures that are 95% white space - I wonder whose DUMB idea that was?)

Actually the way UV maps are used is part of the problem. If you look at M2 or V2 texture maps, you will notice that the eyes represent a very small portion of the overall map. If you want to do a close up of the face and have the eyes look good you need a large texture map just to get the eyes right. The best solution to this problem is not to create extra large UV maps but to use multiple UV maps. When DAZ did V3 they put the eyes on a separate map. (Still I have no idea why we need >1024x1024 maps for eyes). Smart use of multiple UV maps will reduce the need for high res. Unfortunately many designers try to put everything on one map, thus the need for high res maps. One thing that I have done a lot is to scale the texture maps in photoshop. Make a back up of the high res maps incase you need them later. You can make new MAT poses to apply the different res maps. Unless you are doing very close up work, I doubt if you will ever see the difference between a 1kx1k map and a 4kx4k map if the render is 1024x768 or less.

2000x2000 eyelash transmaps are a bit much I agree, even for printwork but you never know how a customer is going to decide to use what they buy. Suppose they decide to do a huge eye, even for the web you might want a 2000x2000 eyelash transmap then! lol I would much rather have a huge version of everything and reduce it myself in PSP then to have the merchants sell lo-res versions of textures. My personal pet peeve is items that call for bump maps when you load them ;) It never fails that I didn't create them and Poser locks up on me. Also makes for more work if I decide to use the item in Poser 5. I would rather have bump maps applied with a separate pose file. lol

FyreSpiryt : Yes that setting is supposed to reduce the textures but I have no idea what the transfer pipeline looks like in Poser 5 and I'm sure the texture load is much slower when it has to reduce. Also, I have no idea how Poser 5 is doing the reduction. In other words, you might do way better reducing the texture in photoshop and cleaning it up vs trusting the FF reduction. As for comments about the large textures when shipped. I do agree that shipping a large texture is probably OK if it is something useful. One can always reduce it. Some cases like fishnet stockings or grates on floors can become problems when reduced. But as was pointed out, a 1500x1500 eye texure for the 3rd gen models is absurd ( unless someone is actually doing renders of eyeballs close up ).

Thanks Ronstuff, My head is swimming a little so lets see if I get this: in practical terms, if one reduces the size (dimensions>size) of a tex. map then it doesn't matter because it will be proportionally applied to the figure? My understanding is, it is kind of like using a towel for a blanket? You would have to use several of them and they wouldn't necessarily line up? By reducing the tex. map size wouldn't that result in less quality? You would be changing the pixel information on a tex. map. Or, does it even matter, if not, then why are tex. maps made so large? My reasoning is you can get more info in a larger file, therefore more quality? I have applied small tex maps to large objects with bad results. Tiling sometimes solves this depending on what kind of effect you are going for; but, mostly it is unatural.

bm

sory but it is not the size of picture only that sometime is a problem... i had two packages from the same merchant... wfor two similar models (mayadolland animedoll. hint :) both texture package had roughly same size of about 2500x2500 or so. one constantly locked when rendered, the other wasn't. when i looked back in m runtime folder, i discovered that one of the texure package was very lightly compressed and the other was better compressed (yet no quality loss) in jpg. for a 2500x2500 imagesize, one of the package displayed textures of about 400kb a piece, and the other somethign like 3.5mb a piece. guess which one was locking my poser ? :) so just recompressed the textures at same size but better algorythme (used jpg new export filter from pixia 2.8, with better quality/size control.. took half quality/half size compromis esettign and checked then after ward on a sample patch of the original texture if any change could be noticable at naked eye and it wasn't :) so one advice would be: compress your texturesz wisely (not too much, but not too loose) and well, yes for parts that are not bound to be used in closeup render, may not need to make them 4000x4000... you can WORK with such file at home when making your textures, but maybe for the end user, unless he has poser 5, 1gb ram and such, some package may b e unusable without serious fix.. and usually someone bought somethign because he wnted not to do it himself, so most of the time he will be upset if he has to fix it. ps: it is no flame toward the merchant whom package i spoke of... in fact, i guess the textures were able to be sued with p5 or better computer than mine so it is my fault; yet well, many people do not have mosnter computer. mine is a gaming one and already decently goes with nowadays games. I would nto expect something like p4 to requires as much horsepower than latest games...

If the size of the compressed file is significant, is a JPEG the best option every time? I'm thinking of transmaps for eyelashes -- how many colours do they use? I'm thinking they're greyscales, so they don't need 24-bit colour. One transmap I've been working on is plain black and white, and I saved it as a 2-colour PNG file. Which works fine in Poser 5. The other advantage over a JPEG for that file is that I don't get those blurred edges. But am I just saving disk space (or download time)? Or does it make rendering a bit less memory-hungry?

Past of the problem with texture maps is that the key objects in Poser are 3D, and some parts can get very close to the camera. So a few pixels on the texture map can be covering a lot of pixels on the render. 512 pixels to map head to toe on a 6-foot man is about 7 pixels per inch. Do a Kitchener-style render of that man pointing at the camera/viewer, and it's not going to be enough. On the other hand, if it's right about memory allocation being in powers of 2, those 3000x3000 hi-res maps for Vicki are a bit silly.

Couldn't agree with you more, mason. All very valid points, and it's about time someone said it. But Riddokun also hit the nail on the head. Texture compression is easily AS IMPORTANT as size. If you make a 2048 x 2048 tex and save it at jpeg 8, it will be WORSE in quality than a 1024 x 1024 at jpeg 12. What's the point in making a huge texture if you use a crap compression factor on it? Here's what I do. All objects mapped properly, ie. NOT laid out neatly side by side on the template, but 'stacked' for maximum resolution (and also, wherever possible, scaled in proportion to each other). All scans done directly from the original materials at 4096 x 4096, then reduced in Photoshop. Maximum tex size (with some rare exceptions) 1024 x 1024, and all texs to the power of 2. All textures saved in Photoshop using 'Save for web' at maximum quality (100). mac

"On the other hand, if it's right about memory allocation being in powers of 2, those 3000x3000 hi-res maps for Vicki are a bit silly." One thing to keep in mind about almost all computers you'll run into currently - ultimately EVERTHING resolves to a power of 2 - it's all about binary. Anything that doesn't itself resolve into a power of 2 will mean a waste of resources at some point. Sometimes you can't help it (we don't operate digitally), but always keep that in the back of your head when working in the digital world. Unzipped

Riddokun- I could only guess at how poser does its tetxure storage but I do know that ultimately, as in any render engine, the texture will have to be unpacked to full size when rasterized in the engine. Now it could be that poser stores the item as a compressed JPG but I'm doubting that highly. There are texture compression techniques that allow compressed textures to be rasterized but that's slower. Now what could be happening is perhaps poser fire fly has a bug in its jpg decompressor that crashes FF on certain jpg types and in only certain situations. That seems more likely than compressing the texture as a smaller jpg.

yes i know tthings need to be unpacked in memory and such.. that why i was surprised too when i witnessed the behavior i just reported. for information, i can add that i never turn on "textured" preview, and so textures only show at render.. i don't know if it was pertinent in this case could you explain a bit abotu rasterized thing ? i guess i never heard abotu that, or with a local transalted word so i do not really know what ir refers to. well anyway it is not fireply on my case, i have p4. and all my crashes went from huge textures (regarding resolution) but not all hugely sized textures did crash it at all, only those textures with both high resolution size and file size :) and that is strange, i admit...

looks at my standard 512 or 1024 size and is happy Agreed, I try to size the map depending on the importance of the object I'm texturing. Somehting important and/or large like a full clothing suit gets a big map, the hat alone gets a smaller one. Seems logical to me. Use material color settings where you can and skip maps entirely. Reflection maps can be small and fuzzy, nobody will ever see. This is also why I appreciate the unimesh map design with the eyes on a separate map. That means I don't get stuck with a huge empty second map just to change eyecolour. My poor lil machine really appreciates that. :) Also uvmappers should think about space usage when laying out maps. Make big things take more space and little things take less space. Double up objects where it makes sense.

interesting thread! thanks for all the info.

Is it ok to make rectangular textures (ie: 512 x 1024 or 2048 x 1024), just as long as the dimensions are still power of 2?

Yep thanks all for your comments! Most instructive :)

lyrra: about the eyes, yes i definitly agree.. you can imagine it gave me some hell in some of my project.. i really consider trying to make a free lam eyeprop (if i had basic notions of modelling and cr2 editing)

so far just made prototypes with primitives out of poser, just to check ;!

that is why for example i preferred working with mayadoll or even v3/new daz figures than with older characters. thats also why i finally tried to groupas much eye mapped textures on a single file not to waste space. (so far i can put v2,m2, steph2, dork, posette, and v3/m3 -iris only for them) on a single map with correct details :)

for for some other items it can be hard for sure

i was also surprised of the amount of good look you could achieve just with material settings (reflexion, highlight and base color) as weel as a lame bumpmap or reflection map :)

an interesting alternative is to work with small/low textures if character seen from far or medium distance, and load/switch a bigger detailed one for close up and portraits for example. this can also apply with the meshes themselves: look at neftis"s Elle character. have a low polygon version, a regular one, and a special separated high resolution head mesh for close up portraits and facial animation.. found it a gound idea

Message edited on: 06/01/2004 18:00

I'm a day late and a dollar short, but I don't think I made myself clear. It's not so much about the raw size for me, as the efficiency. For instance, my problem with the eye texture I mentioned wasn't that it was 1500 x 1500; it's that it was made of solid blocks of color 6 pixels square. It had obviously been a 250 x 250 map that was resized without even so much as a little smoothing.
I'll admit, one of my guilty texture-making pleasures is 1000 x 1000 eyes. They're fun to make, and that's the size the 400 pixel irises I like to make fit onto. And, every once in a while, I do an extreme closeup that benefits from it.

To me, and also echoing what some others have said, it's about making textures as large as you need, but no larger. If you need 4096, that's fine. It can be scaled down if not needed, but there's no way to regain the detail that's not there. But if you only need 512 and make it 4096 anyway, that's wasteful. And ironically, the texture probably won't look as good as if it were 512! Again, efficiency.

I have a couple of questions kind of related, but I don't know if anyone will have the answers. Does the "max texture size" option also cause P5 to scale smaller textures up? That wouldn't make sense to me, but I think I read somewhere that it did, and since when does Poser make sense?

Also, where to procedurals fit in relation? Unless it's a huge difference in memory, I guess it'll depend on your shader versus the map you could use instead, but is there a rule of thumb? When are you better off going to a node-heavy procedural and when better with a map, if you can get the same effect with each?

Most people also render at around 2000x or more & scale down later in an image editor.
Thats not to say I'd have a 2500x for a lamp though,lol...that does sound a little extreme.especially for a reflect map,
Also correct modelling is essential,if you plan on tiling or stacking a texture across a wall then you'll need to cut the wall accordingly,

Cheers
Stefan

Message edited on: 06/01/2004 23:42

Attached Link: Low Res Maps = Beautiful Images...

Message edited on: 06/03/2004 06:49

Oh, on if you do click the link, on page 4 of the thread, the artist showcases the texture map used and the way it's laid out. Thought that may be helpful for folks to see also.

Thanks Teyon for this link! It is very instructive to see what can be done with very low polygon count and clever textures and no space wasted UV maps! Makes me think things quite differently now.

incredible.. i saw the render, the ploly wired mesh peview, and the texture mapping.. incredible. sure i think we may be pushing it too far ourselves with poser i guess :) because the result there is very good and according to our poser standards it is unbelievable must learn !

I suspect that one reason why Poser stuff uses such detailed meshes and textures is that it can get used for closeups. That figure is maybe closer to the (now free) Lo-res Vicky-1. In a game, moving, you wouldn't notice the lower quality. For another example, a cine film projected from 35mm has a huge image from a smaller effective frame than a 35mm still camera. Because it's a moving picture, we don't get a chance to see the film grain.

well i never found lowres vicky 1 for free (lack of cover magazines for it :), and well, of course game engines are aimed to animating and realtime, but well here on this topic we were shown a "still" render picture, and it is still impressive, i say

Attached Link: http://www.renderosity.com/viewed.ez?galleryid=630465&Start=1&Artist=narcissus&ByArtist=Yes

A small example. There are similar catsuit meshes for both Posette and Vicky, and there is a Spacepilot texture available for the Vicky catsuit, from from DAX 3D. It is possible to re-arrange the Vicky texture to work with the Posette UV mapping. The free texture is sized at 3000x3000. To get the pixel size of the actual mesh surface to be the same, the Posette catsuit template (made with UVmapper), had to be 4050x4050. There's quite a few differences at the edges of the mesh, but the end result was OK.