Udim? Texture sets? Texel Density? What is it and why is it? Analysis from A to Z.
Today I want to share with you tips on using Udim, texture sets, texel density, padding, mipmapping and other techniques in various situations and when errors occur.
This material was created by the author himself, but it also includes information obtained from other sources.
Editor and creator of the idea: Daniil Lyapin (lipa_l1).
Thank you for the help in writing the article from wonderful high-level artists: _Shelest_, Titan_Fair, Alexey Yarmola, nikiost, trasnik.
Special thanks:
Dmitry Bogumil — for providing files/screenshots of the monster model.
gRen — for providing screenshots with errors in baking id cards on weapons.
Sites where a small part of the information was taken from:
I would appreciate it if you could help me fix the typos, and if you have any questions, please email: lipindan2003@gmail.com , in telegram — @lipa_l1, in discord — @lipa_l1.
**The article has been translated, and there may be errors in translation.**
**The article has been translated, and there may be errors in translation.**
"All materials on this website are copyrighted (including design). It is prohibited to copy, distribute or otherwise use information and objects without the prior consent of the copyright holder."
The following programs and addon were used to create this article:
* Blender 4.2.0 / Blender 4.3.0;
* UVPackMaster 3.2.2 PRO / UVPackmaster v3.3.0 PRO;
* Marmoset Toolbag 4.06 (4062) / Marmoset Toolbag 5.0 (5008);
* Substance Painter 3D v10.0.1 build 3757.
If you encounter any discrepancies with the actions described in the article when working with these programs, it is recommended to use the most similar versions of these programs to avoid problems.
—————————Udim—————————
U-Dimension, also known as Udim, is a system for optimizing your model by increasing the number of UV tiles, where shells fill their space, increasing in size, which makes it possible not to limit yourself to using high Texel Density (hereinafter referred to as TD), and the texture, in turn, will be displayed better quality.
Udim is called in different ways, but all these terms are synonymous. Don't be alarmed if you find different names in different sources: Udim, Udim scheme, Udim system, tiles, UV pieces, squares on UV, tiles, and so on.
The goal of Udim is to optimize the process of using high textures. It depends on the density of the TD. Usually, beginners use only one UV tile to scan, bake texture maps, and create the texture itself. *The exception is small—area models.
Over time, the understanding of three-dimensional graphics grows, opening the way to complex technical issues of the model itself. This is where many people encounter difficulties when creating 3D models with textures created in Udim.
Why is TD so important?
Texel Density is a value that shows how many pixels of the future texture will be displayed on 1 centimeter of the model's surface. *Another unit of measurement is possible: m, mm.
It is important to know how TD works, many are disoriented in this understanding, because you can find quite a lot of articles and blogs on this topic, everyone is trying to tell only a couple percent of the information, due to the lack of accurate knowledge.
TD is directly related to the distance to the model, the further away we are from our object, the less we can see it, thus the smaller TD is needed, and the closer, the better the flaws will be visible and, accordingly, a higher TD is needed. Hence the question, which TD should I use?
To begin with, it is impossible to calculate the rational number of UV tiles, TD and resolution of future textures in our head, we need a starting point that will serve as a beginning. Yes, you can use a large number of high-resolution tiles, but this will not make sense if you waste resources and poor performance, so it is important to choose the best combination, and due to the difference in shape, surface area and size of the models, you cannot treat all objects equally, an individual approach is needed. TD will serve as our starting point.
How do I get the required TD? Here you need the distance to the desired object and the value of the lens distortion (the focus value), there is a formula for calculating the value. We also need reference values, again, you can't take the value out of your head, you need tests to find the best value or take ready-made values from other artists' projects as a reference.
Formula:
If we substitute the value, the formula will be reduced to the following form:
We are looking at the focus value.
Now our model: according to the formula, TD came out 86.106, we substitute this value in the Texel Density addon:
where TD is the desired texel, px/cm;
TDe — texel of the standard, px/cm;
L is the distance from the camera to the surface of the model, m;
Lэ is the distance from the camera to the surface of the reference model, m;
F is the focal length of the camera, mm;
Fэ is the focal length of the camera at the reference, mm.
The resolution of the texture is adjusted to the TD we have already selected, which we have calculated, the TD value remains unchanged, and based on it, you make a decision on the resolution of the texture and the number of UV tiles.
How to use the standard correctly.
You have found a successful render that looks both beautiful and optimized. Next, we need to find out from this part of the model what the TD is, the distance and focus of the camera to the surface (the nearest point on the surface).
If we substitute the value, the formula will be reduced to the following form:
Where 1.34 is the value that you got when shortening the formula. Next, you can have several of these values in your bookmarks for future projects, as excellent examples that may focus on quality, optimization, or something in between quality and sufficient optimization. And then it is only necessary to substitute the distance to the surface of the model and the focus value of the camera in subsequent actions. *The value 1.34 is fictional.
How to measure the distance correctly?
In Blender, the ruler works perfectly, all you need to do is connect the camera point and the nearest object surface.
We are looking at the focus value.
We have a distance of 1.60266 meters and a lens distortion of 120 m, after adjusting the size and applying transformations with modifiers.
There is a value that you can use as a reference, it perfectly shows the model and is qualitatively optimized.
1.15 — this value is applied on a motorcycle.
Now our model: according to the formula, TD came out 86.106, we substitute this value in the Texel Density addon:
The shells must be prepared before use. *Overlapping should be done, the shells should be straightened as much as possible, have no stretch marks and scaled depending on the visibility of the parts of the model.
If this value is applied to our sample, we get the following image. Using the Uvatoolkit addon, I'm visually showing you the result of our calculations.
If you increase the texture resolution without changing the TD, the UV space will only scale, which will provoke the use of more or less UV tiles, because the shells remain the same, without changing in size. The TD that we have calculated does not change under any circumstances, the resolution is already adjusting to it, Udim will be needed as support.
Not many people realize that neither the size of the model is important, nor the resolution of the texture, but only the distance and distortion of the lens. The size of the model does affect the TD, but the distance to it will be different when rendering or in games, and it compensates for the size of the model.
*The calculations themselves must be done from the moment of the closest object visibility distance during rendering, including the closest moment in the game.
*Overlaps should be removed when creating a UV scan with the required TD value. They will be calculated as extra area values, which we definitely do not need, so we remove them and after all the manipulations we return them to the stacks.
Here are the steps for correct operation:
1) Set the correct model size, apply all modifiers and transformations. We clean the model of unnecessary curves and other debris.
2) We do a scan with overlaps and reduce the shells of the barely noticeable parts of the model.
3) We calculate using the formula, knowing the closest distance to the model, the distance from the camera to the model, and the camera focus value. We take the necessary option as a standard.
4) We give the scan the desired TD, then select the best option from the texture resolution and Udim. (2k and 3 uv tiles, 4k and 1 uv tile).
Frequently Asked questions and Answers:
1) When you need to make a render very close to the model, and it is large, the same rules apply here, the distance is calculated, a reference is taken and the desired TD is calculated, if the render is on one side or the object in the game will not be visible from all other sides, then in the scan we can use only those shells that They will be visible in most cases, and the rest will be reduced at your discretion, so we will be able to use low-resolution textures, which must be done before calculating the formula.
We reduce the shells that we don't need, which makes the visible parts of the object more detailed, as a rule, this is the base that you should know, and the UV scan should be already prepared, the formula only gives the value for which we select the future texture resolution using Udim.
2) It is important to have a model size equal to the objects in our reality, applied transformations and the absence of active modifiers, as well as the absence of internal polygons, excess debris in the form of curves and polygons on our model.
3) According to the logic of such calculations, it is not difficult to understand that the smaller the model, the fewer the shells, and the fewer the shells in the UV space, the worse the texture will look. From this we can understand that when rendering a full-screen model, let's say it is small, then we will need to increase the shells to increase TD, with larger model sizes, the shells will be larger, which means they will need to be reduced on the contrary, because TD will most likely be needed less.
4) The TD that we received is already a ready-made value, it does not need to be changed anymore, the texture is already selected for it using Udim. Depending on the situation, you can use either a large texture resolution or a smaller one.
5) The "trash can" model will rarely be seen in the game, but it is possible to approach it and examine it closely, then it is worth making calculations from this distance. For optimization, there is LOD technology and MipMapping, so you don't have to worry about the high quality of the model up close, it will be different in the distance.
6) When creating a model for the game, it is worth considering the distance and the focus value, in each game, depending on the character's face and other factors, the distance may be different, as well as the field of view, it is worth considering the maximum field of view, it also affects the distance, in addition to the usual distance. What is not worth considering is the approximation using third-party programs and objects in the game itself (for example, the SVD sight).
7) We have an animation in which actions take place in an ordinary room and at a certain moment a mosquito flies in front of the screen, what kind of TD should it have? Don't be alarmed, we take the moment in front of the camera as the closest and use the distance, the focus value and the reference value to find the TD we need, even considering that most of the time the mosquito itself will be far from the camera.
Cases when we have a frame (we set the correct size of the model, apply all modifiers and transformations. We ALWAYS clean the model of unnecessary curves and other debris).
If there are TD limits, then use the formula to select the correct td values, then reduce or increase the shells.
If there are texture frames, then we use the formula to select the correct td values, using the shell size and Udim, we select the necessary values for resolution.
If there is a Udim frame, then we use the formula to select the correct td values and change the resolution that will match the Udim, as well as scale the shells.
If the frames are based on td and texture resolution, then we use the formula to select the correct td values and, using Udim, scale the shells by changing td.
If the resolution limits are Udim, then we use the formula to select the correct td values, check how td shows itself with Udim and the required resolution, and scale the shells if necessary by changing td.
If the frames are for Udim and td, then use the formula to select the correct td values, then reduce or increase the shells and select the desired textures for Udim.
A framework with all the limitations — there may only be a suggestion to resize the model or maximize the UV. Including changing the range to the model itself.
With large frames for the model, in rare cases, scale the model itself so that it is further away from the camera.
*Scaling of shells should start from the barely noticeable parts of the model to the noticeable ones.
Before texturing, you can use texture checkers. They are usually embedded in 3D modeling programs or available as additional plug-ins (addons). You can also use a regular texture to see blurring and artifacts during the final scan with TD already applied.
For example, I use texture checker from the UVToolkit addon in Blender 4.3.
In order to enable Udim in Blender, follow the instructions from the screenshots:
This technology helps to optimize the game and make it smoother. When the player is far away from a certain object in the game world, the texture of that object becomes worse. This is because the weight of the texture is reduced, and the game works more efficiently.
In this illustration, one texture seems to flow into another. The texture changes where the cuts were made on the model. It looks something like this:
2. On the left in the middle, select the desired color (we chose green) and click on the FillObject button.
3. As you can see, new material was assigned to the selected part.
To upload, it is enough to select only the first texture in the count (1001), and all the others (by map type), if available, will be uploaded by UE itself. For Udim to work, do not forget to check the Enable virtual texture support box.
In Unreal Engine 5.5, it looks like this.
There are already painted high poly models for baking ID cards.
We get these UV tiles or prepared texture sets for baking.
Next, all our UV tiles should look similar. We are transferring all our tiles to one tile number 1001. Although they are all superimposed on each other, they have a difference in color, so do not be afraid. This is necessary for the Marmoset Toolbag to see them, since it does not see Udim in Multiple Texture Sets. If we have symmetrical parts of the model or copies, then do not forget about overlapping. The Marmoset Toolbag itself (when baking texture sets) uses a tile space, that is, a space that repeats the texture across all tiles, based on which we transfer overlapped shells to the side on any tile, there will be no defects, but they also take part, since baked maps are then automatically superimposed on them, baking it will be based on materials only in 1001, and the maps that we want to see are superimposed on each tile.
Overlapping will look similar:
As a result, there will be such a picture, here you can see that the overlapping shells are superimposed on each other, is this normal?
Here we observe 3 of our materials (2 that we created ourselves and assigned, and 1 that is automatically created on models during 3D modeling).
After setting up the bake section, we start filling the cards and see the following option. The cards are baked well, and you can additionally adjust some settings for a more ideal option.
After baking, do not forget to make all the cards the correct naming for convenient import into SP, this is done by adding .1001 and so on at the end of each card and by card type.
Further, for texturing, we should have neither spaced overlappings nor texture sets, so before switching to SP, we return the overlappings to their places, and we distribute the sets themselves and remove all materials from them. How will we texture when we have some shells on top of each other (overlapping)? Texturing will affect all shells, but where they are on top of each other, our texture will repeat, so they don't overlay everything, otherwise when texturing, say, blood on our model, this blood will be on all parts of the model, so when importing our low poly model, we save the overlay, as we did from the very beginning the beginning. As a result, we have 3 tiles, and each one has overlappings.
This method only creates space so that we can understand where the tiles are separated and scan exactly across the tiles, rather than in empty space, but a good alternative to this is add—ons for creating UV scans on semi—automation: RizomUV, UVLayout and add-ons for the programs themselves, for example, one of the best is UVPackMaster3PRO, which creates excellent scans on different tiles without previously created Udim spaces in Blender.
As you already understood, Udim plays the role of an assistant, with an already known TD, we can choose an 8k texture, it will perfectly match our TD, but the problem is that textures of this quality are quite difficult to find, it has several more disadvantages: optimization in the game, the greater weight of one texture takes longer to load than several textures with lower quality, but at the same time with the same file size, the file size itself also affects, it is much more profitable to use 3 4k textures than one 8k.
Let's look at the advantages of Udim in different situations.:
1) As I wrote above, it's not easy to find high-resolution textures and even more so that they match your desires, it's much easier and better to use several low-resolution textures.
2) From a technical point of view, if we take into account the amount of memory that the texture itself occupies, then using three 2k textures will be more profitable than using one 4k texture (at the same time, our TD will increase even more, if 2.5 textures are needed for our TD, then we take more, 3 textures, shells, respectively, they will become more, which will give an increase in TD):
*Do not forget that as a result of baking, we get not one map, but about five: normal map, thickness and others. And this is only for one UV tile.
2k — 15,941 * 5 cards * 3 UV scans for each card = ~233 mb
4k — 58.143 * 5 without Udim = ~283 mb
8k — 207.953 * 5 without Udim = ~1 GB
—————————————————
Now let's discuss how the models are applied in game development. Udim gives us noticeable advantages here. This approach uses the so–called padding technique, which, apart from game models, is not used in cinema, visualization, medicine and other industries, but there are rare exceptions.
Padding is the distance between the shells. You can see that we have switched from conventional Udim to padding.
3D graphics is not only about creating beautiful models. It uses complex computing technologies that are interconnected with each other.
Padding is related to TD, TD is related to texture resolution, and texture resolution is related to Udim. But why does it also play a role in this article?
This distance is required for all UV tiles of models that are loaded into game engines. The engines have a special optimization technique called MipMapping.
This technology has almost no drawbacks. It not only makes the game more fun, but also improves the image quality on the screen.
This technology helps to optimize the game and make it smoother. When the player is far away from a certain object in the game world, the texture of that object becomes worse. This is because the weight of the texture is reduced, and the game works more efficiently.
The settings in the game engine allow you to control this process.
With poor padding, artifacts can be noticeable — these are unwanted effects that spoil the visual perception of the game.
In this illustration, one texture seems to flow into another. The texture changes where the cuts were made on the model. It looks something like this:
If the indentation is done correctly, the result will be the same as on the third line of images: the colors are mixed with each other "wisely". However, it is not necessary to perform such a "smart" color filling between shells yourself. There are special programs such as Marmoset Toolbag and Substance 3D Painter that can do this automatically. That is why we have maps that are completely covered in color, and not purely based on shells.
But if you want to have a customizable padding, the Marmoset Toolbag has this functionality:
In the image, you can see the gaps between the shells. However, padding is not only the distance between them. It can also be called in different ways: the distance in pixels, the empty space between the shells, and so on. But the point remains the same: if all the shells fit snugly together, the textures will look terrible. This is especially noticeable when there are many shells on the UV tile that are pressed against each other. With large shells, at least there is a small distance between them (due to the complex shapes, the distance forms itself), but that's not the point. It is important to perform the scan correctly.
MipMapping is a technology that is always used, and it is directly related to padding.
The effect of padding on TD:
The simplest example to understand this is when you have a large number of shells, and the padding between them is 16 pixels. The padding itself takes up a lot of space.
When using Udim, we have several UV tiles, and they have the same padding everywhere. This is the advantage of using Udim.
When you have a lot of islands, there should be a padding of 16 pixels around each of them, and the shells can't be bigger. Even if you enlarge one island, the others will become smaller due to padding. But when using Udim, the padding remains the same, and the UV islands may be larger.
This was all about game development, but most models are created for visualization, motion design, cinema and other graphic projects where padding is not required. In such cases, the padding value is set to 1.
These 2 screenshots show the difference: in the filling of space and in the TD indicator with different padding.
I want to share important information about using a small indentation. The problem is as follows:
In this image, the model is covered with a Material ID card. This map allows you to see the problem better than others. However, this problem will occur with all maps, as with further texturing, textures will be displayed incorrectly, especially on glossy surfaces.
The problem is not the size of the padding, but the quality of the baked textures.
We have 1x samples. That's the whole problem. When we adjust the padding, for example, to a resolution of 2k, we have a distance between the shells of 16 px, the program automatically thinks that the quality of texturing will correspond to 2k, in this case, you should not change your views on the quality of baking, if you decide in the future to make textures and have already adjusted the padding, it means that when when baking the cards, you should pay attention to the quality of the baking. Samples are just one puzzle when building a high—quality image, it is worth considering absolutely all puzzles, including resolution, padding, and bit format. There is also an option that only the Material ID is baked in low quality, because it doesn't matter what quality the monochrome color is, but as you already know, it makes sense to bake even such a non-demanding card in good quality.
————————————————————
We are considering the process of baking a color map from a highly polygonal model. Some of you may not have known, but high—poly models are used to bake a Material ID (Color ID), a color map that will be used to texture a low-poly model.
In addition to the Material ID, there are other baking cards such as Group ID, Object ID, UV ID, albedo, and Vertex color.
When we bake the Material ID, Marmoset analyzes the colors and automatically adapts them to a low-poly model so that they are as different as possible on the color palette (to avoid texturing problems). In other words, the colors on a low-poly model will look different than on a high-poly model.
In the image below, you can see what the painted high-poly model looked like. The image above already shows the colors on the low-poly model: pink turned green, blue remained the same, but became a little lighter.
Group ID is a card that cakes colors based on the same properties of several objects (identical materials, rotation angle and other transformations, data (at the model code level), animations, copies (for example, many chairs that differ only in size).
Object ID is a map that bakes colors for each object locally. For example, a house will have 3 colors on the map (roof, floor, and walls), as a simple example.
UV ID — cakes different colors for each island on the UV scan (for a low-poly object, respectively).
Albedo is the same as the Material ID baking option, but only this card is baked with the same colors that you have assigned to the high-poly object.
Vertex Color is a card that needs to be baked using ZBrush (and other programs).:
1. Select a part of the object.
2. On the left in the middle, select the desired color (we chose green) and click on the FillObject button.
3. As you can see, new material was assigned to the selected part.
The principle of operation is similar to other 3D editors.
If you don't see any colors when loading models into the Marmoset Toolbag, and all the shaders look white, then the problem will be as follows:
It is necessary to simply enable the necessary checkboxes in the program settings, in our case it is fbx.
—————————————————
I also want to add information about the distance between the shells. Many people do not understand how this works, and they may make a mistake if they set the padding to 16 px without knowing the rules of mipmapping. As a result, they will lose some of the important space in the tile.
256 = 2 px
512 = 4 px
1024 = 8 px
2048 = 16 px
4096 = 32 px
When setting up padding, it is important to remember that this is the distance inside the tile between the shells. However, the distance from the tile boundary itself is calculated differently.
The bottom line is that the UV space repeats all the time. The end of the right edge of the tile becomes the beginning of the left edge. Therefore, with a padding of 16 px, the distance between the edges and the shells should be twice as small. If the distance from the shell to the right edge is 8 px, then the remaining 8 px will fall on the left edge. In this case, the padding will be performed correctly.
Many designers either do not use side margins, or set them to be the same — for example, 16 pixels each. If you do not use indentation at all, the textures can overlap each other, creating the appearance of seams and low-quality textures.
In UVPackMaster3PRO, the padding setting function looks something like this:
Margin is the distance between shells in the program. In each program, it is calculated according to its own parameters. This method is not suitable for us.
We check the Pixel Margin (pixel distance).
Pixel Margin — the distance between the shells;
Pixel Padding — the distance between the shell and the tile border;
Extra Margin To Others — the distance between selected and unselected shells;
Texture Size is the future resolution of the texture.
To summarize, it is necessary to understand well what padding, td is and how to pack UV correctly. After all, even the slight curvature of the shells and the irrational use of space can affect the number of UV tiles. We also did not address the issue of stretch marks, which sometimes occur when using perfectly smooth and beautiful shells.
Approach the packaging wisely, because it not only affects the appearance of the model, but can also significantly improve game performance by increasing fps 1.3 times. There are hundreds of such models in the scene. The portfolio also looks at your scan.
The images show an example of a correct scan and an incorrect one (to put it mildly).
3) The use of Udim in games / movies — we have already touched on this point a little higher.
We already know the advantages for games, but the essence of this point is that the biggest advantage of using Udim is to use it in: movies, visualizations, concept projects, logos and other images with high detail.
The main difficulty is finding the necessary textures with high resolution. They are not so easy to find or create yourself. Therefore, it is better to split them into many small fragments and use low—resolution textures - this will improve the quality.
Due to the gigantic size of the models, the shells will also have a large size. After calculating the required TD, we will have to pack it all into one UV tile with a resolution of 64k, or use a Udim system that will consist of, say, 20 UV tiles, and each tile will have a resolution of 1k.
Therefore, Udim are always used in fairly large quantities to create giant structures to properly demonstrate the quality of textures. Even in games, this method is used when there can be 20 tiles on the model, and this is normal.
When using a large number of tiles in the cinema, the quality will be as realistic as possible, and there will be no losses, of course, the rendering will be long, but in the cinema these are small things, so in all films there are a lot of Udim tiles on each model, and Udim takes a huge turn to create logos and realistic scenes that are used as pictures.
For an example, you can take a look at the following images:
As you can see, it is almost impossible to achieve such detail without a large number of tiles.
4) Optimization when used in the engine:
This advantage will be mainly useful for the Unreal Engine, since the use of Udim is more developed in it than in Unity.
We will use UE4 as an example. When Udim is loaded, they are used as a single shader (material = shader), each type (group) of maps is a new shader, that is, we will have about 5 main maps (or shaders), but they already consist of Udim tiles. The "VT" label serves as proof of the successful loading of the cards.
To upload, it is enough to select only the first texture in the count (1001), and all the others (by map type), if available, will be uploaded by UE itself. For Udim to work, do not forget to check the Enable virtual texture support box.
In Unreal Engine 5.5, it looks like this.
How is this approach useful? When you split the texture into parts, the model becomes more optimized. This is because Udim has flexibility. When using this method, the texture is not fully loaded, but gradually. It's like opening one 100 GB file and ten 10 GB files. One file will take a long time to download, as it is large and unified. And when using ten files, each of them is processed separately, which allows you to increase the speed of working with similar models. As a result, we get optimization. When you approach the model, the nearest maps are loaded.
5) Many programs support Udim, in addition to those described above, Houdini, Nuke, Modo, Mari, 3D Coat, ZBrush, Maya, 3ds Max, Cinema 4D, Substance Designer, Marvelous Designer and many others work well with Udim.
6) In addition to the fact that Udim can be used to achieve realism on the entire model, it can also be used to prioritize resolution on some parts of the model, for example, if your model looks from afar in the game and in the future there will be a scene with it where only the head is visible, then it is not necessary to do the entire a model with 8k textures, since it will take up extra space and require more resources, it is enough to make the entire model of good quality, and make the head, which will be shown in close-up in the future, of high quality.
7) Using Udim in Marmoset Toolbag 5.
After the long-awaited Marmoset Toolbag 5 update, you can bake texture maps without bothering with texture sets, and bake using the Udim tiles themselves.
*When using overlapping, it is important to do it efficiently, otherwise you can expect problems with shading, this is due to a very sensitive setting in the bake — Fix Mirrored Tangents section. You didn't have to use it for texture sets.
*When using overlapping, it is important to do it efficiently, otherwise you can expect problems with shading, this is due to a very sensitive setting in the bake — Fix Mirrored Tangents section. You didn't have to use it for texture sets.
8) The convenience of Udim in Substance 3D Painter (hereinafter referred to as SP). * It will be described below.
———————Texture sets———————
Many people confuse them with Udim.
They have 3 formulations.
1) This is one of the texturing functions in Substance 3D Painter or specially prepared Udim for Substance 3D Painter.
2) Specially prepared Udim for Marmoset Toolbag.
3) Udim is so confused because of the similarity, because texture sets are groups of textures.
Let's take a closer look at them.
1) This is one of the texturing functions in Substance 3D Painter or specially prepared Udim for Substance 3D Painter.
The Substance 3D Painter program is unique in that it supports both Udim and texture sets, but it is important to understand that texture sets are not a different arrangement of shells, tiles, etc., but specially prepared Udim (it is assumed that these are the same Udim, but have their own specifics, namely color shells (or parts of the model that consist of shells) and are painted exclusively by the low poly model).
Udim for Substance 3D Painter is the most convenient technology, unlike non—close texture sets, which are not effective and may interfere with texturing.
It will be quite difficult to tell this due to the large number of combinations of texture sets with Udim, so an experiment was conducted on the most rational way of texturing, using Udim, texture sets or not using them at all.
https://vk.cc/cH0uAw - a link to the experiment itself. (You can leave comments in the project itself, I will answer them).
As you can see, each of the 3 methods has at least one worker, but Udim gets the first place for its uniqueness and all the amenities, I advise you to use it in all your projects, even with 1 tile.
The screenshot shows an item using the Udim system.
Texture sets are very inconvenient, in my own experience I have not seen anyone who would use them for texturing, the result is that you should not even paint your low poly model for SP, and if you did, it would be better to remove the assigned materials.
2) Specially prepared Udim for Marmoset Toolbag.
Before the release of version 5 of the Marmoset Toolbag, texture sets became very popular due to their important functionality — baking maps according to certain tiles.
Here we have painted low poly models or created texture sets.
There are already painted high poly models for baking ID cards.
We get these UV tiles or prepared texture sets for baking.
Next, all our UV tiles should look similar. We are transferring all our tiles to one tile number 1001. Although they are all superimposed on each other, they have a difference in color, so do not be afraid. This is necessary for the Marmoset Toolbag to see them, since it does not see Udim in Multiple Texture Sets. If we have symmetrical parts of the model or copies, then do not forget about overlapping. The Marmoset Toolbag itself (when baking texture sets) uses a tile space, that is, a space that repeats the texture across all tiles, based on which we transfer overlapped shells to the side on any tile, there will be no defects, but they also take part, since baked maps are then automatically superimposed on them, baking it will be based on materials only in 1001, and the maps that we want to see are superimposed on each tile.
Overlapping will look similar:
As a result, there will be such a picture, here you can see that the overlapping shells are superimposed on each other, is this normal?
I will answer you "Yes". Each texture set of a certain part of the model has its own overlapping, it is impossible for the shells of one set to intersect with each other, but we have 3 sets, and here that the shells of 1001 intersect with each other, which is 1002, this is normal, we see this intersection, but not the program. She sees the colors by which she starts baking, and overlapping is needed to prevent defects during baking and optimize for quality improvement.
Such defects may occur if overlapping is ignored (when the shells are at 1001 bids).
Shading is where it shouldn't be.
If you don't do the overlapping too well, there may be problems with texturing in the future.
For a better understanding, I advise you to read articles on this topic.
Go to the Marmoset Toolbag program and start baking.
Select Multiple Texture Sets.
Here we observe 3 of our materials (2 that we created ourselves and assigned, and 1 that is automatically created on models during 3D modeling).
After setting up the bake section, we start filling the cards and see the following option. The cards are baked well, and you can additionally adjust some settings for a more ideal option.
After baking, do not forget to make all the cards the correct naming for convenient import into SP, this is done by adding .1001 and so on at the end of each card and by card type.
Further, for texturing, we should have neither spaced overlappings nor texture sets, so before switching to SP, we return the overlappings to their places, and we distribute the sets themselves and remove all materials from them. How will we texture when we have some shells on top of each other (overlapping)? Texturing will affect all shells, but where they are on top of each other, our texture will repeat, so they don't overlay everything, otherwise when texturing, say, blood on our model, this blood will be on all parts of the model, so when importing our low poly model, we save the overlay, as we did from the very beginning the beginning. As a result, we have 3 tiles, and each one has overlappings.
3) Udim is so confused because of the similarity, because texture sets are groups of textures.
Due to the artists' limited experience, incorrect translation, and fake articles, discussions on this issue are beginning, and they are very loud.
Even the well-known GPT chat cannot give a clear answer, as the society itself has created a lot of false information on the Internet on this topic.
I shared the information that I personally have from my own experience and that of other artists, confirming it with photo arguments, I hope you will not confuse these terms in the future.
Thank you for reading the article, a lot of effort was spent on it, as I want to tell you everything in a clear language, but as you can see, Udim touches on many other topics that cause difficulty in explanation. Even before the article was published, I received a pretty great feedback, which motivates me to do even more for you. You definitely haven't wasted 70 minutes of your time. 
Replies
First of all, if you're going to post on an art forum, it's a bit rude to ask people to email any feedback. We all learn from each other when discussions occur in full view. It's also easy to turn on email notifications for forum responses.
Secondly, you shouldn't claim copyright on all this content because a lot of it is not yours. You should ask for permission before posting someone else's content (unless there's a clear license allowing it already), and you should include a credit line for each image. It's just the right thing to do.
It would also be nice to have a concise summary of your points.
And like Eric wrote, it's common practice to credit every image or at least create a footnote so people don't have to piece together that information from the links themselves.
Then, your article takes quite an authorative tone ("people often don't understand this and that, this information is seldomly shared" etc.). I'll give you the benefit of the doubt and assume that there might be less information available in Russian, but in English, there is plenty of good information around. Perhaps this was only meant as a comprehensive guide, your workflow that works for you and your understanding of the information you have found, but the tone still matters, especially if people find that the result can't live up to what's promised. A positive side effect might be that you'll get more feedback that way ("someone is wrong on the internet!"), but the whole article is so long that people might just give up.
Based on what I've read, the article is in urgent need of an editor (who isn't yourself). Some of that might be due to the translation (and at least the technical terms should be correct, e.g. "focal length" isn't "lens distortion"), but some sentences and transitions seem to be empty filler at best and some information doesn't make sense or is misleading at worst.
UDIM is ultimately just a practical and standardized way to organize multiple UV sets in tiles and many of the advantages you ascribe to it are just advantages of using multiple textures. The alternatives aren't to use UDIM or use one big texture. And some arguments, like "it's hard to find high resolution textures" have nothing to do with the question if people use UDIM or not or how the textures are applied. It's still the same source texture and resolution, whether it fills one UDIM or a part of one big texture.
Also, what's your target audience? Many might find the introduction with the formula for keeping TD consistent to a reference render (the image resolution of which would matter, btw., and should match the target image resolution at least roughly) a bit overwhelming, especially if they are more artsy than technical types. This is something that can easily be determined by eye in many cases, and no deleting of overlapping shells etc. is necessary either for that. So for me, it would make more sense to put that in a later chapter or an appendix or even separate article for people who want to automate this in some form or create stricter project wide standards or who simply feel lost without an exact method.
In my opinion, all that's really needed that early into the article is just a definition of what TD is and what needs to be considered when determining it (without an existing TD reference).
This probably sounds a bit harsh and some of it might not apply to the original Russian text. I do see the appeal to have all of this in one article, especially with application specific instructions, and I certainly appreciate the effort, but a proper English version would need lots of work in my opinion. I'm not the most technically minded person myself, so I might be wrong about some of this, but while this article might bring more clarity for some, I'm afraid it's overall quite confusing in its current state. I didn't read everything, and there's lots of good information in there, too, but I also often thought "Yeah... but...", "Kinda... but ...", "What?", but it's just too much to go through everything in detail.
Thank you for your detailed comment, at the moment, as you said, we need someone who will do the correct translation, I have found such a person and now we would like to discuss how to do it better. I am the author of this article and I agree that it is overloaded, but it's good that you didn't see it initially)