What is Texture Padding and How to Master it 

Thanks for watching the video, I'm glad you found it useful.

I'm glad you enjoyed the video, it's unfortunate these topics weren't covered in your class.

3d art schools are a scam i did the same shit.

Hey thanks a lot, new content every two weeks. Make sure you watch this if you liked that ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-5e6zvJqVqlA.html

You're welcome, thanks for leaving a comment.

Thanks, glad you found what you were looking for.

Thanks glad you found the video useful. Thanks for subscribing.

Thanks, I try to read every comment and respond to any questions. I'm still actively developing the Mel Script Mega Pack, but I haven't done RU-vid in a while. I'm hoping to make some new videos one day, but right now I can't seem to find the time.

@@malcolm341 scripts is some high level thing for me :D I try to make my way in 3D, just 4,5 years in, but the only problem that I don't have motivation and willpower, so I can't ask for good money because I can't be responsible if I can't trust myself) Sad :D One day I want to try to grow my social media like you do with youtube

@@malcolm341 man that's actually really funny. Now with new knowledge I noticed that Minecraft doesn't have mipmaps in vanilla and now I can't unsee it :D You created a professional deformation for me :D I found a setting to change it with a mod, but this is mindblowing how top 1 game in the world by sales doesn't work like every single game out there :D It just build different 8-)

@@sergeyandreev2599 I've never tried Minecraft, but maybe the creators didn't like the look of how mips would blur the textures, but I think even for stylized games they're valuable if the game can run them at 8x or 16x anisotropic texture filtering.

Thanks a lot, glad you're finding the channel useful. I like to understand how things work under the hood so when things go wrong I know where to look.

I'm glad you enjoyed the video, thanks for leaving a comment.

That's great, thanks for leaving a comment and letting me know.

Yeah, ps1 was the worst, but ps2 had quite a bit as well.

Thanks, man, always appreciate your positive feedback.

Thanks, feel free to share.

That's awesome, thanks for you comment.

That's great, thanks for letting me know.

Thanks for your comment.

Great thanks for leaving a comment and letting me know you liked the video.

@@malcolm341 haha yur welcme,people like u r the ones who truly helps us out when we having queries.and no one's to answer us. 👍

Thanks for letting me know, glad you found it useful.

That's great, I'm glad you found it useful.

That's great, thanks for your positive feedback.

Thanks, glad you found it useful.

Thanks a lot, glad you found it useful.

Thanks for commenting and letting me know it was useful.

Thanks I'm glad you enjoyed the video. I believe mip mapping even at 1x1 pixels speeds up rendering, while omitting the 1x1 texture would save memory it would save so little memory compared to the performance gain you get, that being said an optimization we've often done on games I've worked on is to turn off mips completely on lightmaps and sky boxes because they're always lower than the average texel density anyways and that can save a substantial amount of memory.

Thanks. If you're just starting out you can get a free student copy of Maya, 3DS Max, or the full version of Blender, Unreal, or Unity. You can practice at home by reading the help docs or looking for beginner tutorials on RU-vid. Facebook is also a good place to post questions by joining a Maya beginners group. Besides that polycount.com is a good forum to post questions and look at other people's work. Another route is to go to 3D art school, or you can take online course from a website like pluralsight, kind of depends on how you want to work and how much or how little you want to spend. I'd recommend trying the free stuff first and if you love doing it think about trying some online courses.

Thanks for commenting, I'm glad you found the video useful.

Thanks for commenting and letting me know.

I'm glad you enjoyed the video, thanks for commenting.

Awesome, thanks.

Thanks very much for that positive feedback.

Thanks a lot. Yes manually packing is always better, there's no way that I know of in Maya to auto-pad it after you've packed it manually. What I do is measure two shells and align them with the padding I like and then use that as a visual reference to eyeball the distance for the rest of the shells. Not a great solution, but that's how I've seen most people in studios do it, at least that will prevent you from accidentally padding out by large amounts.

@@malcolm341 Ok thanks for the advice. That's pretty much how I've been doing it so good to know I'm not missing something. One thing I've been doing to help with the eyeballing is to divide up the grid in the UV editor so each square roughly represents the amount of padding you want. I'd really like an option to manually pack and then add padding after. Or set a padding value and have islands snap so they will always keep that distance from other islands. I haven't found any software that does this though so I guess for now it's eyeballing or autopacking.

@@justanothernobody7142 Good idea, in the past I've made a grid texture that a checker every eight pixels and use pixel snapping to align them manually. I've thought about how to solve this problem, but it's quite tricky because if the total overall shells get too big they then need to be scaled down or shuffled around to refit defeating what we want. I know a way to scale each shell up by a certain amount, but this could cause overlap. I think this program may be able to do it, but I found it kind of complex for the simple games objects I unwrap www.rizom-lab.com/products/rizomuv-vs/

Thanks, in the default layout tool you could set the border padding to 4, that would give you 4 on each side which would be 8 total for the border, and then for the padding you could set it to 8. Or if you want to be a bit more aggressive you could go padding 10 and border 5 which would be 10 and 10.

Thanks.

Thanks for commenting and letting me know you liked the video.

@@malcolm341 Super helpful honestly, thanks again👍

Never played it, I've been told I have a deep voice.

If you scaled horizontally in this case to fill the empty space you'd have quite a lot of extra resolution horizontally in one direction which can be hard to texture across or at low resolutions can look strange sometimes. If you're texturing in Photoshop it will be really hard to texture, in Substance Painter it's not that big of a deal because of projection correction. Fill the wasted space if it looks good, but a better solution may be to use a rectangular texture if the unwrap is mostly rectangular.

Great question, yes correct if you bake 4k at 16 padding that will scale down as it mips to 2k and 8pad, then 1k and 4pad. That would be the safest way to do it, in practice we've found in games you can pad less than that so it's generally up to the individual artist to test out and get comfortable or the project will have some established guidelines to follow. In film I've been told everything is 4k UDIM tiles and they pad 32 just to be safe because texture memory is not at a premium like games.

@@malcolm341 Thank you for the clarification of my doubt, Malcom. What you said about the films I had no idea, thank you for that too. Great content, keep it up.

I believe you'll run into texture bleed in any application that uses mip mapping. Severity of bleed will vary depending on texture resolution and number of samples of anisotropic filtering used in the shader to display the mip maps. I've never worked in TV or film so I can't speak to that, but the test case of the grey cube with the red edges I used in this video does appear in the Maya viewport when the texture node is set to mipmap, and also in the arnold render. So I'd image this will issue will also show up in most/all DCC apps as well as substance painter.

Most people manually place the shells after achieving the target texel density. It's not a good practice to rely on the layout tool to achieve the padding for games because as you mentioned it will shrink the shells. It's just a painful manual process, typically how I work is after I achieve target texel density and do a rough manual layout to confirm all my shells will fit in 0-1 (excluding tiling parts of course), then I manually try to place them in a square layout and eyeball the padding. I measure the first two shells using the measure tool so then I can visually reference the distance for the other shells. After everything is spaced and packed into a square I select all the shells and move the pivot to the bottom of the square and uniformly scale them up together until they take up as much space as possible. It's really a lot of annoying manual labour with a lot of back and forth as you go I haven't found a great trick to doing this more procedurally.

If you go to file import and import the scene with the materials in it those materials will be available in your currently open scene. To avoid clutter you could apply your materials to a polygon object and then export just the model with file export selection.

1. Just the space is enough to stop the bleeding, making the space between the atlases transparent would actually be dangerous and may lead to transparency bleeding. 2. 8-10 pixels of space between UV shells is usually fine for most projects I've worked on when a 2k map is used. For 4k maps it would be 16-20 pixels of space. I haven't worked with Substance Designer, but if the maps were baked in it they should have padding built in. If you imported the textures from another program into Designer there's a filter you can get which will pad the images for you inside Designer. This thread has a link to the Substance Designer padding filter at the bottom. forum.substance3d.com/index.php/topic,1876.msg8502/highlight,substance+designer+padding.html#msg8502 if that doesn't work you can always manually dilate them using a plugin or the xNormal Photoshop filter. 3. All map types should be dilated the same, they will all have bleeding issues, but in slightly different ways, the base colour is usually the worst, but sometimes normal and roughness can look bad as well.

@@malcolm341 Wow man, that was so quick! Thank you very much for helping me and even giving me more information such as the link with the plugin for Substance or Photoshop filter. I hope I won't encounter any further problems with my project. I wish you good luck with the development of your youtube channel, because you do such a good job. I will recommend your channel to every 3d artist I know in case they may encounter trouble with their projects, problems they might solve by watching one of your video tutorials. Keep up the good work!

Thanks a lot, good luck with your project.

You can do it as the last step before you start creating your texture.

That's a nice trick, thanks for sharing.

Make sure you have the latest service pack, UV operations will sometimes crash Maya. I'm using Maya 2020.4 Also try it on a cube and see if the issue is your specific model or just Maya in general.

Great question, I use 8 for shell and 4 for tile if possible, which would give you 8 for shell and 8 for tile because the tile doubles up on the sides.

@@malcolm341 ah i get it now! thank you so much for the reply, malcolm, appreciate it :)

Yes pretty much, great question. With any texture size there will always be texture blending because eventually the texture will map down to 2x2 and then 1x1. In practice with 4x to 8x anisotropic texture filtering 8 pixels seems to be a good middle ground even though it zeros out at 256 like you showed. I've worked with people that use 4 pix padding on 2k and there stuff still looked fine in games. If you want to be really safe you could go with 10 pixels padding. It's an inexact science.

Thanks glad you found it useful.

In a 4k technically 16 pixels so when the texture mip maps down to 2k you have 8 pixels. I've found that if you pad 8 pixels with the Maya built in tools using the 2k option, then export the UV snapshot at 4k you'll still end up with 16 though because you've doubled the texture resolution if that makes sense. So just make sure you test the UV snapshot before you commit to it and measure the distance in Photoshop first. It's kind of weird so I always just set mine to 8 pixels and use the 2k texture option from the menu, then if I save the UV snapshot it's still correct for 4k.

@@malcolm341 cool thank for the info

Glad you found it useful, thanks for leaving a comment.

@@malcolm341 your video is always useful for everyone from beginner to expert because some of us are forgetting something and need to refresh the knowledge in production this is a reminder for all level.

Yes I believe it does.

I think that's the only way to do it. What part isn't working for you?

@@malcolm341 you know there is a function called distribute in rizom , i want to do the same in Maya, well maya will layout shells according to the p value we provide, but the result sometimes won't be precise like 8.0pixels or something like that, maya layout tool sometimes will ignore those shells close to the padding value like 7.5, the padding will not be set to 8, I have searched, but did not find any useful answers, and so, i need to use the maya rizom bridge to switch in between.

@@StrongmanLi-pz7sl I think Maya's layout UVs will always be a little bit inaccurate, like it might land between 2 pixels or something, probably the higher padding you give it the more accurate it becomes.

@@malcolm341 thank you, Maya probably works like that.

On certain objects that are too big or have too many UV shells overlap is the only way to achieve high texel density. The other option is to use a larger texture 4k, or you can use more than one texture to solve the issue. It is dependent on the model what's the best solution to use though. Can you post an image of the model you're working with.

@@malcolm341really thankful. Wow very fast reply. on the same scene can we use both 2k and 4k texture ?? pixel density and padding both is important or we can use only pixel density technique ?? How to send you pic ?? On mail ?? Thank you sir

Google drive would work if you post the link here, or if you feel like it just share the model on google drive and I'll take a look. You can technically use as many textures as you want on a model, but in games we try to keep this to a minimum. You can also use a 4k texture and a 2k texture on the same model and have the same texel density if you set it up right, but again this is not that common. It really depends on what you're making and how big it is. If I could see the model I'd be able to give you some better advise.

@@malcolm341 sir i have a problem with my model so i will send you later. So will pls consider my request. Thank you sir.

What if you use material layers in UE4 shader, big padding still needed?

Interesting question, the amount to pad is somewhat subjective, I've been using 8pix padding for a long time on 2048x2048 textures, but I've met people that swear by 10pix and even a person that uses 3pix which seems risky to me, but it seemed to work in engine. As the textures get smaller in your case a 1024x1024 technically speaking I would use a 4pix padding on it, that being said if 1024x1024 is the max texture size for the top level mip I'd test this in game first to make sure there aren't any errors, this will also depend on the amount of anisotropic filtering your in game shader has to blend between the mip textures. For the panoramic I would always use the rule of thumb of going with the bigger dimension as the padding so I'd pad 8pix just to be on the safe side.

@@malcolm341 ok gotcha, I almost never use 1Ks. In Any case I believe that 4px should be the absolute minimum for any case scenarios (512s) Thank you Malcom and have a good day!

@@odeca8121 Thanks for commenting, fun question.

Hi oXY nary, no nothing fancy going on to my knowledge, using Unreal and sometimes we only pad to 6 pixels, but 8 is safer I find. It could be that your anisotropic filtering is lower than 4x on whatever shader you're working with. Never had to go as high as 16 or 32, If you're needing to go that high it could be that your shader is using trilinear filtering rather than anisotropic filtering, or perhaps you're using solid colours for a more stylized texture style, but 8 pixels should work fine for most textures. If you have a copy of toolbag you can import your model there and under the material settings you can change the filtering type and see if that gives you different results as a test.

@@malcolm341 Thanks for the reply Malcolm. Indeed I am using more stylized coloring, so there isn't as much noise at the edges to bleed. I think I found a good writeup by Earthquake (Joe Wilson), that kinda gives whats going on. As per where I had issues is exactly where he describes (Atlas maps with harsh angles). polycount.com/discussion/comment/2576424/#Comment_2576424 . In short, I think 8 works for your type of assets and workflow, but I guess for what I'm doing it depends.

In Maya technically applying maps is not required just the material on the faces you want will get into Unreal, then you plug the maps into the material in Unreal and apply those to the mesh in the element slots.

@@malcolm341 do you have a video suggestion to see this process, I also have trouble get maps to show properly after importing from substance to Marmoset.

@@Musiclover6404 here's the Marmoset basic workflow ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-IIfCNeZcTwU.html

Thanks a lot for your purchase and supporting me I really appreciate it. 70% is actually a low bar, I would definitely try to hit that always and try to get to 80%. Having a minimum value to hit is really important because it keeps the assets consistent between artists, we had 40+ external environment artists working on Gears 5 from different teams and countries and them delivering below 70% really wasn't acceptable for what the consumer expected from that game in terms of visual quality. As well if you set a minimum of 70% this makes the decision making process around when to use addition textures and tricks to keep the texel density high much easier. If you say "just make it look good enough" everything starts to look much worse than expected and people start doing crazy things like putting 4k textures on tiny props because they aren't doing good UVs and are trying to hide it with large textures. If you have to hit 70% it also gives the reviewer an objective way to determine if the prop looks good and gives the artist clear direction where they need to improve before they move on to the texturing stage. Draw calls and texture usage will always be a limiting factor no matter what generation we're on, as processing power increases so does the visual quality requirements so you just lose that new power and get better visuals, you never get any free stuff and you always have to worry about how many textures are currently streamed in and how many drawcalls are visible. If you look at all the PS5 games currently in development they all run at 30fps which is a clear indicator they had to cut frame rate to get the nice visuals, on XboxSX you'll see some games at 60fps because they're pushing that feature, but not all. So what we already know about next gen is it's better, but not a huge leap, it will probably be visually the same as a current gen PC with all the same performance problems. What is super impressive about next gen is the low cost you'll be able to buy the console for compared to buying those some components in a PC. UE5 looks cool, but it has the same constraints as UE4. You're still bound by drawcalls, how many textures are currently streamed in and everything else you need to draw postfx etc. In the Ninite demo you can clearly see they have no real game running and the walking demo is running at 30fps, that means they have no headroom for anything else, what if the game required more than one character on screen, or some AI, can't do it in UE5 too slow, so what they demoed can't be used to make a game currently. I would think of the UE5 demo as look what might be possible one day, we're all going to be really disappointed if we think all AAA games are going to look like this on ps5 when the console launches. I'm not a fan of Epic's marketing they've done this fake promise before in the last two console generations, on Xbox360 it was real time lighting for Gears 1, never happened. For XboxOne it was real time GI, SVOGI, never happened. For Xbox Series X it's Lumen and Ninite and based on their current demo at 30fps with no game running I would imagine that won't happen either. That would be super cool if I was wrong, but I just don't see it with the next gen hardware specs. A good example is the Halo Infinite demo, that's running on the most powerful console in the world, but I don't think that's the best looking visuals I've seen so they clearly have a lot of challenges getting it to look good even with the more powerful hardware. Thanks for your question, have a great day.

@@malcolm341 Thank you for such an in-depth response and explaining the thought process behind the 70% rule. Additionally, you make a lot of good points about next-gen as well. However, does your rule refer to individual props taking up 70? Or multiple objects in a scene taking up 70? Personally for me when it comes to a single prop that warrants its own UV set, needing to hit at least 70, that is typically not an issue since that prop is most likely going to be large and complex and then it becomes more about using less space than finding ways to use more space haha. Though what happens when that prop has hit its texel density goal but there's still space left over? The problem I run into is when fitting multiple assets in a scene into a single UV set. Because assets are of different shape and size and the number of objects your working with, to begin with, also plays a factor. I find it is not always possible to get everything to fit so perfectly and use every last drop of space due to the case by case nature of UVs in that situation. Either way, this has influenced my thoughts and I will aim for 70 in the future. Thanks again for your explanation.

@@askinganime6023 That's a great question. I was referring to a single prop needing to take at least 70%. In Unreal for objects that are too big and you simply can't get to texel density no matter how tight you pack it, there are a couple options: 1. Use more than one texture (increases draw calls) 2. Use trim sheets and stacked UVs (increases draw calls) For modular buildings we generally used 2-3 textures, one tiling texture one trim sheet, one unique unwrapped details. Then we'd try to reuse those materials on the whole modular set where possible. Drawcalls were strictly controlled on Gears 5 because we shipped the game at 4k 60fps on Xbox OneX which was challening, and also a slap in the face to these next gen games running at 4k 30fps on faster hardward then we had. We would definitely could not have done it without being super efficient and rigid with our art rules. Most things were modular on Gears 5 so we didn't really have a lot of situations were objects were combined to one texture page. To optimize later everything went into the HLOD tool for combing and automatic LOD creation which had a whole team of people working on tuning the HLODS to prevent pop in when you played the game If you hit 70% and there's still space left over we found scaling up the tiny details by 50% was a good way to fill in the extra space and actually looked better to have higher texel density on small details, this was perceptional and was an interesting rule we developed. For oddly shaped assets we would also use rectangular textures occasionally if the UVs fit better that way, but most of the time we just shuffled the UVs around until we got to 70%. If it was really not possible then we'd look into using a trim sheet or other kind of texture to help make it more efficient.