How to get hi-res bakes on 1024 and 512 images?

[IIianII]

The title says it all.

I always get blurry textures even when I bake it on a larger image, soon as I downsize it to 1024 or 512 depending on size of the object I get a blurry result.

Sometimes it messes up the image

What should I do?

[arton Rotaru]

There are only that many pixels in a texture at a certain size. If you downsize a texture you are going to lose a huge amount of color information of your original texture. That's why it's often recommened to bake Ambient Occlusion maps at twice of the desired size, to blur jagged edges when downsized.

Smallish, detailed patterns will usually be destroyed when downsized, though

Making the most of the UV mapping is key to get a reasonable resolution in 512's or 1k textures. If the objects are too big, seamless tiling textures are best suited.

Do you have an example texture (or part of)?

Edited January 11, 2018 by arton Rotaru

[Rolig Loon]

35 minutes ago, arton Rotaru said:

Smallish, detailed patterns will usually be destroyed when downsized, though

That's especially true for fine details, like lines that are only one or two pixels wide. Those will either become very faint due to averaging with adjacent pixels or will disappear completely. It's a good idea to ask whether you really need a lot of detail before you go to all the trouble of creating and uploading a texture. Sometimes detail really is important, but much of the time people will only glance at a texture briefly -- or not at all if it is too complex for their graphics card to render quickly.

[ChinRey]

Between them Arton and Rolig sum up the essentials of good SL texturing exceptionally well.

But to add one thing: have you considered the various methods for scaling? Most image editors offer you a list of options how all those pixels are to be merged into fewer ones.

Here are two examples from paint.net and Gimp:

                    

Other programs will probably have different options or use different names for them.

I'm not going to go into details about things I have no clue about myself but what I can say is that the choice of resampling/interpolation method can have a huge effect on the result and if you're not happy with the first option, it's well worth trying the other alternatives.

Edited January 11, 2018 by ChinRey

[OptimoMaximo]

The best way to preserve your high res texture is to work with  multiple UVSets, where the first is a highres image with all UVs laid within the UV space, the second one instead uses the ssame UVs split by material. This way you can work highres on a single image using a UVSet, to then bake onto the second UV set's UVs. The two sets of UVs stay separate and unrelated from each other in regard of the editing you do, so you basically might also have your second UVSet done as 4 lighmaps. The baking will do the map conversion for you

Here's an example: a cylinder, UVs covering from 0 to 1, 1 UV tile

This belongs to the default UVSet, now i will copy this UVSet into a new one and make some adjustments

When baking, you can designate any texture mapped onto the first UVSet to bake its shaded results onto any UVset you want. Once you've got all done, you can delete the first UVSet and all of your transfer/remap textures work exactly as they did on the original. Cutting more shells or rearranging the UV entirely will NOT affect the transfer and seams will be preserved. So, say you had a 2K texture for the first UVSet, you can reasonably let the software do its job on multiple images in order to retain the number of pixels that each UV shell had in the original texture.

In this case, since the original UVs on the cylinder were badly made, the texel density is higher on the second UVset now. If i wanted 100% texel density preservation, i would have sampled it at 2048 resolution, then set it at same value on all shells in the second UVset, sampling it to a 1024 texture size. Cutting and slicing would most likely occur to split exceedingly large shells to fit within multiple UV ranges (which i like to call "tiles" for sake of simplicity). After the image baking is done, I will later use the UVs in these tiles to assign materials for SL to split the texturable faces on the imported model. Nothing prevents you from using tileable textures as the bake input, your second UVSet will output a matching resolution baked image

[Penny Patton]

Another important issue that many content creators in SL overlook is that having a high-res texture ends up being a waste if the texture map itself is full of empty or hidden space. It feels like the majority of textures I've seen in SL use only a tiny fraction of of the pixels, the rest of the file just eating up valuable VRAM for no reason.

Also, look for ways you can reuse parts of the texture across your model. I've seen mesh chains that give every single link in the chain pixel space in the texture when they could totally have a single instance of the chain link texture used for every link in the chain. Doing that would allow for both higher resolution even if the whole texture is a fraction of the size.

[OptimoMaximo]

2 hours ago, Penny Patton said:

Another important issue that many content creators in SL overlook is that having a high-res texture ends up being a waste if the texture map itself is full of empty or hidden space.

This is what my method addresses. The UVs aspect ratios wouldn't even matter anymore IF nobody else is supposed to put hands on the texturing after the release. That is one point of optimization of the textures to take all the available space, with unordered/unoriented UVs and perhaps also their planarization if necessary. Things that may make modding a bit overly complicated. If the aim is to allow modding, instead, the UVs should be thought as fewer, very well oriented and well placed shells to make sure others can make quality and optimized content. In both cases, my method helps address the creative work where you may want to be at ease working with a very very hires texture that, for seamless texturing workflow, unfortunately waste quite a bit of space, to THEN think to a SL texturing solution, with more or less compatibility for modding activity, as i was saying just above.

2 hours ago, Penny Patton said:

Also, look for ways you can reuse parts of the texture across your model. I've seen mesh chains that give every single link in the chain pixel space in the texture when they could totally have a single instance of the chain link texture used for every link in the chain.

This is totally true, if the model's textures do not use any baked information in lighting. This was true in games of the past, where the assets were painted, using a 512 texture for a MIRRORED full character (so half the UVs where just ovelapped and flipped to reuse the other side). SL isn't that different, with the classic avatar mirrored arms. This also implied to use planar mapping like crazy, which may work well for flat shapes like walls, but it's definitely bad for rounded shapes as it creates distortions on the sides facing away from the mapping direction.

It is possible to do some work on realistic looking stuff to address the UV overlap, but baked lighting information is really not easy to handle this way, ambient occlusion and reflections can get quite off on some instances. These content creators give the customers what they ask for, or at least they listen to the major requests they do: realism, doesn't matter the cost. I agree with you for what concerns the texture space use, a lot of works could have a better UV mapping for sure. But you can't make a case for all, in this regard, it's the single creator's choice in the end, and overlapping UVs for baked lighting is no go. The example of a chain might be improved for baked lighting, putting, say, 8 links in a UV map, then reusing those to construct the chain and then deform it to go in place, but that requires either a) single object with one material per 8 links section or b) each 8 links section of the chain to be a separate object. Which might be less convenient or efficient.

I've been reading your thread about being against no mod stuff, and your remark here just links to that kind of perspective. The problem isn't the fact that i may or may not agree with you in such regard. The problem sits in the attitude you show against methods different than yours. Hence, if you're so dissatisfied by how 3D modelers do their UV mapping, make your own models with the good UVs you're looking for. I for one will not sell my wearable stuff as modifiable because you can "improve" it. You just can't. Even with mod permissions, you wouldn't be able to pull a crap out of my stuff because, to optimize it to the max, i use my mapping methods that aren't suitable for PS texture painting modding. My full perms stuff instead addresses this need and i make the UVs as easy to use as i possibly can, straightening the UVs. I agree with you, again, that there are many who don't care and make UVs that kill the alive and resurrect the deads. But the need to "improve" other creators' stuff is something you feel and think is happening, while someone else may say you just downgraded the whole look. The point is, instead of mocking people for the mod/no-mod choices, stop buying this stuff and begin to make your own, if you haven't already started. When you got it all and found a solution that serves all possible cases of UV mapping needs and workflows, come back here and teach us all how to achieve the best UVs ever from the height of your expertise. Until then, try to address the scale issue that starts from the inhuman avatar size that the majority of users have. If you get a tiny avatar, you value your money the most because this way your regular sim is 16 times (relatively perceived as) bigger, smaller items that eat up waaaay less LI and better land management/profit by splitting land in even more parcels for a reduced price for the customer. Let's see how many will turn tiny for you.

Edited January 18, 2018 by OptimoMaximo
typo

[Penny Patton]

11 hours ago, OptimoMaximo said:

This is what my method addresses.

Apologies, I was not criticizing your method. I was posting before bed and had not read all of the follow up posts thoroughly. It's simply that I see so many textures in SL wastefully using up VRAM with blank space and redundant pixels so I believe this point needs to be highlighted more often.

11 hours ago, OptimoMaximo said:

But you can't make a case for all, in this regard, it's the single creator's choice in the end, and overlapping UVs for baked lighting is no go.

Arton brought up the fact that repeating patterns can often lose detail, creating a small but high-res and repeatable portion of the pattern is a great way to get around this.

Again, I am merely pointing out the practical issues content creators need to consider. We can debate the merits of artistic freedom until we're blue in the face, but how one textures does affect framerates and lag in measurable ways that have nothing to do with how we feel about it. There's leeway to be had, but people need to be aware of the impact of their texturing choices. That is all I'm trying to say.

11 hours ago, OptimoMaximo said:

I for one will not sell my wearable stuff as modifiable because you can "improve" it. You just can't.

With all due respect, you don't know that. I'm not saying you HAVE to sell mod for my sake, only that I will not buy no-mod myself and that I will share my reasons for this so that others can make informed purchasing decisions based on that information and their own values.

11 hours ago, OptimoMaximo said:

Let's see how many will turn tiny for you.

 I've been in SL since 2005 and talking about the scale issue the whole time. In the years since there has been a massive scaling down of avatars and content in many parts of the grid the likes of which people used to say would never happen. However, if you want to debate these points I would suggest we take it to private messages or that other thread so we don't derail this topic any further. 

[OptimoMaximo]

4 hours ago, Penny Patton said:

With all due respect, you don't know that. I'm not saying you HAVE to sell mod for my sake, only that I will not buy no-mod myself and that I will share my reasons for this so that others can make informed purchasing decisions based on that information and their own values.

Oh yes that i can know you can't mod my stuff, as i make a jumbled mess, with no UV shells connected, many many pieces put in random order. you, as anyone else myself included, can NOT possibly retexture it by hand. That's why you can't. And that's why that stuff is no mod. And i write "COPY ONLY" on the vendor texture, in a font size as big as a colonial house. However, after a demo purchase and a full purchase, bad reviews arrive because "everything should be moddable, give me mod item or i don't delete the bad review". NO. You aren't ABLE to texture at my same level and you would crappify my item. Let alone the fact that i use a mapping method which is a impossible mess to work manually, but even if i did make PS or GIMP usable UVs, you won't make my textures any better. Guaranteed. That's why i use a texture change hud, with color grading options. 

 

4 hours ago, Penny Patton said:

Arton brought up the fact that repeating patterns can often lose detail, creating a small but high-res and repeatable portion of the pattern is a great way to get around this.

The problem with this assumption of yours sits in the UV mapping technique where you downscale the image and loss of details occurs, so you resort to tiling. My method preserves the texel density from any mapping resolution to the same texel density in multiple 1024 maps resolution. I don't think i need to explain to you what texel density is, do I?

The method you describe was used at the time of games like Torchlight/Torchlight2 and even earlier, WoW, i'm sorry. These repeatable patterns are, again, best used when placed in the same coordinates of different UV spaces/tiles, so that you can bake model information on another texture/material/mesh part, not always on the same. What you describe can work, but it doesn't give the user the amount of control you should get on the texture, needed for detailing. Like the chain example, every link can't have the same textures as the shine and ambient occlusion can work well on a few, on the others they would look very off. Should every single link/groups of links get a material on its own to load in another texture? That isn't very optimal, as it calls another set of textures to draw.

I'm not saying that you're full wrong on the whole line, i want to clarify. Or that you're a low level texture creator that can't possibly reach my heigths. I'm saying that you're making assumptions on one method (yours) being best over all the others, when i clearly showed something that not many know it's possible and how to do it. I could make a whole car in one UV tile to make a render out of it, who cares: tiling textures go well there as the lighting is being rendered after a click. Who cares, in such case, of the UV space taken by each shell. Then i upload it an sell it to you as moddable. you're not getting a good result from that because you either tile textures (bleah) or paint detailed ones that have small texture space. Overlapping pieces works in rendering, how about the doors when baking or texture painting, if they're overlapped? Mirrored details to be optimistic. What about moving the mirrored half in another UV tile? you get another material/SL face to texture, a mean to bake to texture only once and apply the same texture (flipped if you also flip the mirrored UV map to have their normals in the correct orientations. Yes, UVs have normals as well.) or bake independently into two different textures.

The fact is that most likely the no mod creator has made their product in a way that works optimally in their mind to get the visual look they planned, which is not guaranteed to work for another person. Or like me, it would be IMPOSSIBLE to texture the item manually because of the mapping method. I couldn't do that myself on my own model, i have to take the texturing stage Maya file again and bake again to the SL maps when done. But rest assured that if i started high res with a 2K map where only a 1024**2 area was wasted space, the final SL mapping gets me 3 fully used texture maps, no resolution loss and no resource loss. This is why you can't generalize about no mod items being a money throwaway and that ALL can be optimized/improved. On a product like mine, the only optimization you could do is to downscale my textures. But then you lose the detail that i so cleverly managed to retain, thanks so much.

[Penny Patton]

1 hour ago, OptimoMaximo said:

That's why you can't.

I'm not going to derail this thread by debating this here. PM me or post in that other thread if you want me to give examples in how I could still make use of mod perms with such creations.

1 hour ago, OptimoMaximo said:

I don't think i need to explain to you what texel density is, do I?

Let's keep it civil. There are multiple ways to tackle this. The way I suggested allows for lower VRAM consumption. Despite your claim, this method is still utilized frequently in modern gaming. Let's also remember that excessive texture use is an even larger issue than it is in most videogames for several key reasons, here's a few of them:

• All content in SL must be downloaded. That includes every single texture. The more you have to download the longer it takes and the more latency spikes you'll experience.
• You also have a limited local cache, when that fills up textures are deleted and have to be redownloaded resulting in yet more lag and more texture thrashing.
• Content in SL is not made by a single individual or team, so textures are not recycled often between various content. SL is frequently throwing far more unique textures at you than any game, modern or otherwise.
• SL does not fully utilize graphics hardware or memory, and what memory it does use must be shared with all of SL's graphics shaders and effects. So even if you have 8 gigs of videocard memory you don't want to throw 8 gigs of textures at it.

 

[OptimoMaximo]

38 minutes ago, Penny Patton said:

Despite your claim, this method is still utilized frequently in modern gaming.

My claim is for pre-rendered textures based games, like SL. If we're to talk about game engines, shading doesn't belong to the color texture at all so it would be perfectly fine. Texture atlases are fine with overlapping UVs, again there's no AO or shine on the color texture. This is the difference you seem to be not grasping. There is people who doesn't WANT to use such techniques because they want their items to have a defined look in their shading, not a painted one with shading flaws here and there. If you read what i wrote carefully, i said i agree with you that some creators make bad UVs coverages and/or split stuff into too many individual textures, but it's what they are demanded by the average customers, who don't even CARE about VRam and what it is

Again, "doesn't matter the cost" and "life-like" are two of the major keypoints in all requests. Failing to meet this expectation means your customers' exodus toward the "higher quality" competitor brand. I recently have written about this "high quality" issue, in regard of mesh bodies, in another thread, so i won't be repeating myself here.

1 hour ago, Penny Patton said:

Content in SL is not made by a single individual or team, so textures are not recycled often between various content. SL is frequently throwing far more unique textures at you than any game, modern or otherwise.

Of course there aren't recycled textures, even if you recycle them in your software, the final rendering is a unique texture for the very nature of pre-rendered (baked) environmental items textures. It seems to me that you're referring to square tileable textures to apply flat on a building surface, in which case you're missing the point of shading. Making a vignette-like shade on the textures edges is not baked shading, nor even gets close to it. I'm sure you can achieve great results with your skills, yet it's not the same and you can't expect designers to agree with your method. So if Vram is a concern, why not looking also at avatar render costs and mesh polygon densities? A png image at 1024 takes on average <= 1 MB, while a collada file containing a high density rigged model can esily get to a few tens megabytes. It all gets compressed in a binary format, still the fewer data in input, the smaller the converted file. Since SL content, as you said, is not made by one individual or a coordinated team, you can't claim a common standard because that complies to your workflow. Are you better at optimizing content? Make tutorial videos and provide a link to those no-mod creators that you have been so dissatisfied with so far. And make them public, so more people would see the method in action and learn. A video like "SL item improvement through modding", where you show the before, the work you do, and the final result. Let the audience decide whether it's an actual improvement, in spite of the fact that you certainly get better viewer rendering stats.

 

[Penny Patton]

16 minutes ago, OptimoMaximo said:

but it's what they are demanded by the average customers, who don't even CARE about VRam and what it is

I've seen this claim bandied about, but I've never seen any evidence to support it. Well optimized content frequently looks as good or better than content with excessive texture use to the average SL user. I've even optimized other creator's work for them and shown them the results. Creators who have made this exact claim. They had no idea I'd altered anything until I pointed it out to them.

16 minutes ago, OptimoMaximo said:

So if Vram is a concern, why not looking also at avatar render costs and mesh polygon densities?

"Whataboutism" is never a good argument. I frequently, and publicly, discuss render weight and polygon density, but neither is what this thread is about. I also share work I've created and provide tips others can use, none of this has anything to do with the current thread and I've told you I'd rather not derail this topic.

16 minutes ago, OptimoMaximo said:

Let the audience decide whether it's an actual improvement, in spite of the fact that you certainly get better viewer rendering stats.

Which is why we share information and ideas for public discussion. How do you expect the audience to decide what they feel is best if you keep information from them?

ETA: I've already been spending more time than I'd like in one thread arguing about optimization and modding content, I'd really rather not get into another. I said all I wanted to say in my first post.

Edited January 19, 2018 by Penny Patton

[ChinRey]

3 hours ago, OptimoMaximo said:

...

but even if i did make PS or GIMP usable UVs, you won't make my textures any better. Guaranteed.

The question there though is how do you define "better" in this context? Better for what? Better for who?

[OptimoMaximo]

9 hours ago, ChinRey said:

The question there though is how do you define "better" in this context? Better for what? Better for who?

Can be that it's impossible to retexture manually for the above mentioned reasons, remapping to random placements and rotations across multiple materials to gain best texture space. splitting with this method will make impossible to avoid seams or any reasoned detail placement, therefore you can't make it any better. 

a few images to explain a bit better

high res texturing, you can definitely mod on this UV. Lot of texture space is unused though

remapping for SL. i split all in several shells across multiple materials

See now why you can't make mods on UVs like this

This was a quick and dirty layout (five tiles) i could make a better coverage on each tile, i admit. However, see how connected, adjacent patches are put in different tiles (materials) and with random orientations. How would you go for modding an item with such UVs? It's gonna be full of seams and it's gonna require a lot of trial and error, for a result that, honestly, won't look as good as the original.

 

As per Penny, i've already spent enough time trying to explain something she doesn't want to hear, so i will just pass and not waste any more time.

[Klytyna]

On 18 January 2018 at 4:11 PM, Penny Patton said:

It feels like the majority of textures I've seen in SL use only a tiny fraction of of the pixels, the rest of the file just eating up valuable VRAM for no reason.

This isn't an SL problem, it's a Hobbyist 3D problem, going back well over a decade and a half.

"Cookie Cutter UV Mapping"

Time was... A lot of uv maps made by hobbyists were done using 'auto uv bake failure' buttons in apps like UVMapper.

In another thread somebody posted their sample texture for a wooden door with a metal doorknob and hinges, they made such a pigs ear of the uv map that out of the 1024 x 1024 texture, the wood for the door took up the equivalent of say 60% - 70% of a 512 x 512, while the ends of the two door knobs got more space than the wooden door it's self.

There's a full perm template mesh maker on the MP, who does a choker made of ribbon... the UV map on a 1024 x 1024 texture would actually use approximately 800 x 50 of that texture map, the rest is unused.

People can't seem to understand that splitting uv maps by material, and using tiled uv maps where possible, and MANUALLY rearranging parts of the map, so that you use 90% of THREE 512 x 512 textures for a door instead of 40% of a 1024 x 1024 is more efficient and gives a higher PPP (Pixels per Polygon).
 

[Codex Alpha]

I like this debate/discussion, I'm learning a lot he he.

I think I've already done what Optimo has done in the past, by accident to solve a 'newb' problem I had. Just trying to process exactly what is going on and the possibilities of it. I was probably already doing it, but was told it was the 'wrong way', and that 'tiling' or inworld was preferred - While I walk around inworld and see amazingly textured with nice shadows and such... feel like noone wants the competition in SL, so instead of competing on workmanship or creativity, they take the tools and knowledge away from newcomers.. Still discussions like this are valuable for those of us apparently not  in the in-crowd. Thanks Optimo for the info.

I would prefer to use multiple textures, and use UV space much more effectively, but I've also heard the argument that using multiple textures can sometimes add up to MORE MB overall, and that the calls themselves to the database also creates overhead..

I'm also one for spending time optimizing the mesh, lods, and textures... also as a work of art if it isn't just a blatant rip off of a real world item.. there is pride and time and work into it.. and it's an insult that someone DEMANDS mod on it, but I've said my peace on that already.

I'll be incorporating what I've read into my newer works now.. knowing I was on the right track all along. Thanks.

 

PS. Whoops. Necro'd.. but still relevant today

 

Edited January 10, 2019 by entity0x

[Kyrah Abattoir]

On 1/20/2018 at 2:51 AM, OptimoMaximo said:

Can be that it's impossible to retexture manually for the above mentioned reasons, remapping to random placements and rotations across multiple materials to gain best texture space. splitting with this method will make impossible to avoid seams or any reasoned detail placement, therefore you can't make it any better. 

a few images to explain a bit better

high res texturing, you can definitely mod on this UV. Lot of texture space is unused though

remapping for SL. i split all in several shells across multiple materials

See now why you can't make mods on UVs like this

This was a quick and dirty layout (five tiles) i could make a better coverage on each tile, i admit. However, see how connected, adjacent patches are put in different tiles (materials) and with random orientations. How would you go for modding an item with such UVs? It's gonna be full of seams and it's gonna require a lot of trial and error, for a result that, honestly, won't look as good as the original.

 

As per Penny, i've already spent enough time trying to explain something she doesn't want to hear, so i will just pass and not waste any more time.

JESUS How many 1024 textures are you using for a boot?

And you absolutely can mod textures like that just fine, texture artists have painted seamless texture seams long before texture baking came to be.

At least your poly density and packing are decent but you should learn to use shared uvs for details that are identical and to straighten your UV patches to avoid cutting pixels in diagonal where you can.

 

As for what your customers "demand" they don't really demand anything. Whether you do a good or a bad job at your model doesn't really matter for them as long as the result is there.

Don't claim it's the customer when it's your own insecurities.

[OptimoMaximo]

1 hour ago, Kyrah Abattoir said:

JESUS How many 1024 textures are you using for a boot?

Who said I actually did that way in the final product? I'm showing my method and did as many of them as could fit in a screencapture to prove the point at hand. 

 

1 hour ago, Kyrah Abattoir said:

And you absolutely can mod textures like that just fine

As long as you can paint on the model, yes. Mudbox for one manages UV Tiles just fine. Doing it planarly in Photoshop or similar can't give good results with my method, since the final model also gets squared-out UVShells, not those you see there, for even better texture space usage, reason why I do a transfer bake from one UVset to another. Manual painting can't get the UV distortion sorted, while the baking does.

Edited January 16, 2019 by OptimoMaximo

[OptimoMaximo]

1 hour ago, Kyrah Abattoir said:

As for what your customers "demand" they don't really demand anything. Whether you do a good or a bad job at your model doesn't really matter for them as long as the result is there.

Don't claim it's the customer when it's your own insecurities.

Oh and for the records: I make stuff for a brand. If the owner would ever ask me to make a 4096 texture for a product and turn that into the equivalent 16*1024, I would do it. I get my payments IF i abide to the request.

[Kyrah Abattoir]

52 minutes ago, OptimoMaximo said:

Oh and for the records: I make stuff for a brand. If the owner would ever ask me to make a 4096 texture for a product and turn that into the equivalent 16*1024, I would do it. I get my payments IF i abide to the request.

Out of curiousity, why don't you cut out the middleman?

[OptimoMaximo]

3 hours ago, Kyrah Abattoir said:

Out of curiousity, why don't you cut out the middleman?

Because i don't want to deal with marketplace (packing, vendor, listing, description, etc) and customers

 

[Christhiana]

Something to keep in mind when unwrapping. Every UV seam you make in blender gets split in the uploader resulting in more polygons.

[Kyrah Abattoir]

4 hours ago, Christhiana said:

Something to keep in mind when unwrapping. Every UV seam you make in blender gets split in the uploader resulting in more polygons.

vertexes.

[Christhiana]

8 hours ago, Kyrah Abattoir said:

vertexes.

Ofcourse...thanks for the correction!

[Kyrah Abattoir]

No problem.

But yeah it won't lead to more polygons however it does induces a slight slowdown on the rendering end because as far as the GPU is concerned, each separate material is considered a separate rendering job. In addition, the more UV seams & hard edges you have the more "constrained" some of the low level rendering optimizations ends up.

It doesn't truly matter tho unless you go to the extreme of splitting every single triangle, and even then it may or may not be noticeable.