How to tile photogrammetry based PBR materials
Вставка
- Опубліковано 4 лип 2018
- #seamremoval #tiling #photogrammetry #scanning #pbr #workflow
Video tutorial which covers my process of PBR environment texture creation using photogrammetry.
This time with the main focus on tiling part I covered:
- building lowpoly for baking in ZBrush to support seam removal
- manual tiling and seam removal in Painter
As an example I picked two, quite hard to tile ground urban surfaces and managed to turn them into nice and clean PBR materials
For those interested into getting more details, here is a link to my photogrammetry guide:
gum.co/YanD
Thanks to those who bought it already, as it really helps me to move forward with my research and create even more content.
Feel free to follow me on Artstation:
www.artstation.com/gbaran
Amazing tutorial, thanks you so much !
Thanks ToKaRiN :), happy you like it
As much as i like this system I think procedural is the way to go. In the time it takes you to scan two different materials and have them game ready a texture artist could of made 1 of those into a procedural material. I do think scan is worth the time though for things like landscape materials it makes a huge difference. Great work ! Thanks for the video :) p.S. now i know why all the scanners on art station are scanning 16k haha
Hey, thanks, I would say.. there is no the best and the only way to create materials. There are many ways to create them. The best way is to mix techniques.. this is why it is worth to know them all. There is also many techniques regarding the reality capture.. its not just a photogrammetry.. but to name a few the most important these are also AI based single image based reconstruction or photometric stereo. Also procedural creation works way better when combined with traditional approach.. or mixed with photogrammetry. I am planning to cover it all in a future in my videos. Hope that makes sense.
Regarding to photogrammetry... I always do a 'reality check' to score materials. Imo procedural material is well made when its not easy to tell it proceduraly generated. It means that usually you need a lot of time to create a fully procedural high quality material which looks photorealistic. To do that you usually need a very skilled material artist. Before you make a material you also need to study it .. gather surface references, learn how the surface responds to the light changes.. gather and introduce correct color values ... understand the material structure ...and at the end introduce all those findings into material. When you capture the reality .. by scanning it ... you get very high surface quality very easily. Also it is way easier to maintain the material consistency across the material library this way when you have a team of material artists to manage.
So regarding just the reality capture techniques.. photogrammetry is cool.. but its an option. Works great for biomes.. for vegetation I would use photometric stereo technique... for simple and non complex surfaces single Image based reconstruction. For generic surfaces like .. painted wall or metals procedural approach works very well etc.
Hope that makes sense :), cheers!
This will be perfect for my thesis!
Hope in a good way at least ;P
@@GrzegorzBaranArt indeed!
I study surveying/topography
The thesis consists in making a very realistic scenario of a building (and this is good material to know how to treat the textures)
@@Acampandoconfrikis sounds great, good luck with your thesis then :)
Just what I was looking for tyvm!
Happy to hear you found it useful :)
Yeah I purchased the pdf as well keep up the good work
Thanks David :D
@Grzegorz you can do it faster with point classes selection (rectangular) second option and so after you build the dense cloud, you bake two meshes in knald or X normal, photoshop/gimp/affinity might be better than cloning, your recorded action will work for other maps, however cool idea with uv plane
Dzięki wielkie :)
np :)
You're doing a very good job with your textures, great attention to detail. I had a question which I was interested in for a while now, if I remember correctly you were working for Ubisoft, right? Would you mind telling me what file format you use in most scenarios for your texture maps which are used in games, and what resolution? I've read somewhere on the internet before that targa would probably be my best bet. As for resolution, people have stated that using 4k textures is usually only worth it for viewmodel textures such as hands, gloves, weapons, etc. And for the rest, it's usually 1024 and 2048.
Right, but unfortunately due to NDA I cant. Hopefully if you pay more attention to certain parts of my videos you should find the answer you are looking for. Regarding to the file format I can tell you that everything depends on your needs and the tech you use. Just read the documentation related to the game engine of your choice and bare in mind when you create materials/textures that color depth changes, resolution changes.. works just one way only. Hope that helps :)
Grzegorz Baran Thanks for your quick response! I see, that's unfortunate. Didn't know that was something that falls under your NDA. I've already learned a lot of very helpful info from your videos, it's very interesting to see how professional environment artists do their job. Do you by any chance have something like a public discord server that you post in often, maybe where you share previews of your personal projects and such? Would be very nice to have an opportunity to get more advice from you there.
@@XsizztEdits Thanks man, really good to hear that. I have a profile on Artstation where I share my results www.artstation.com/gbaran . If you go to material section you should find a lot of additional info for each since I always share as much info as possible about how they were made: www.artstation.com/gbaran/albums/86576
I also run a blog on Artstation about my experience with the HDRI: www.artstation.com/gbaran/blog and also trying to help everyone who contacts me on Artstations built in communicator.
... and finally I run this youtube channel where I always try to help everyone who needs my help in comments section.
I believe those videos are the most beneficial way to share my knowledge and experience but also since it takes a lot of effort and time to make one I really struggle with time.
Dunno would it make sense to setup a Diskord channel :) .. might try if I find that more people find it beneficial and useful to them :)
Cheers!
Thanks so much for doing the tutorial. I have a problem when baking with the substance designer that I couldn't bake the Color Map for some reason. Error message: Baking failed (Color Map from Mesh)
Invalid input data. Check the logs for more information. Could you please guide me to fix this? I have exported the vertex color from the Reality capture.
Thanks, sure. Everything depends where color information is being stored. You can store color information using vertices and its called 'vertex color' but you can store it in a separate bitmap (which is called a texture). You can even store color information in both at the same time. In your case apparently there is not any color information stored as a 'vertex color' information. Why? I have no clue, maybe you use a format which doesnt support vertex color information? I would suggest to use an FBX or PLY. Both support vertex color information. As a different solution I would suggest to try export color information as a texture and select this texture as a color information source during baking. Feel free to check my recent videos where I use the texture to carry color data and explain why :). Hope that helps. Cheers!
Great tutorial! but I have one problem. The clone tool seems inconsistent for me. I crate the paint layer for the Clone tool and set everything to passthrough. When I look back at the normal channel the Y Direction for the normal seems to be flipped. When I bake out the texture for the normal it's very apparent that the cloning flipped in the normal map. Any suggestions on how to fix this?
Sorry, cant help you.. I guess it is a bug.. or you set something incorrectly. Could you please try to turn the height map off as see does it affect the normalmap?
@@GrzegorzBaranArt Hahah, 2 years too late, but it's probably some DirectX/OpenGL combo box option for the normal map. =)
Your guide has helped me a lot, thank you :-D Btw, the clone tool seems inconsistent for me. Sometimes it will not clone the area I've sampled and sometimes it will. Have you experienced this bug as well? Although the tool seems to work perfectly for non-plane meshes.
Happy to hear that David. About your problem.. you need to make sure Paint layer is selected when cloning. Second: make sure brush is big enough to fill the space.. sometimes when you have gradient on it and it is too small, cloned data is outside of its range. To solve it you can grab sharp one without soft edge transition. Third, make sure area you are cloning from is inside the canvas, sometimes when it is not it has no data to clone from. Fourth idea I have is that you can try to increase the canvas resolution.. Hope that helps. Cheers!
@@GrzegorzBaranArt Tyvm!
The low mesh you used in Painter is the same one you used (created and modified) in ZBrush? I guess UVs needs to be there for the mesh plane. Trying to learn myself to use this technique/workflow, thank you for all the tutorials and info!
Nope. In painter I use different mesh - the same over and over :). It needs to get perfect uv coverage as it is made to preview tiling. I created two types of it, on is a simple plane with UV 1:1 and another one covers 20% more to help me with the edges - this one has a bit tricky uvmapping since painter was working fine just within 1 UDIM- you can see it at 1:15 of this tutorial preview: ua-cam.com/video/kRO8tToVMgg/v-deo.html
I have added these files to my tutorials in payed version but they - especially the simple plane with basic UVmap - are easy to make really.
Hope that helps :)
I was confused because I saw you using a low poly plane in ZBrush and moving his vertices to align better to the texture (making edges not perfectly straight). The most clear part for me is that we use the plane we created in ZBrush with UVs to bake the textures from the high poly model to the low one. The one with 20% more space is a second low poly plane we create (in any 3d software) with his own UVs too? thank you
Yeah, the plane in Zbrush is the canvas to cover area we want to capture. It doesnt need to be filled in 100% with data and edges doesnt matter much as long they are not important for tiling later (like brick texture). The most important part on that stage with UV mapping is to have it consistent without any stretching.. not to cover 100$ of the uv space. So every lowpoly has to be dedicated to highpoly and each one might be different to another.
The plane in Substance Painter has different functionality, it is used for seam removal preview really - tiling (seam removal), rearrange the texture, main bug fixing. So it needs UV space filled in 100% as it represents final texture coverage along the final bitmap.
So these are two different planes. Can you use just one for it? Yes, for sure, but as long as it fills full UV space and UV is not modified for baking :) but it wouldnt be efficient as .. what if you cover tree bark which is not flat? Or tricky rocky surface? Cheers!
Gotcha, I'll make some test, thank you!
Are you makeing photos of pedestrian on 90% straight angle or you little bit bend your camera while makeing photo ?
It doesnt matter as soon you have all data covered and no information gaps left. Of course bigger angle causes depth of field issues so areas out of focal point lose sharpness with distance so it has to be compensited with smaller aperture and longer exposure time. But as long you dont shot with the flash where light falloff would be obvious, angle doesnt matter too much. So if I can and its not a big deal, I usually shot perpendicular to the surface. If for any reason I cant - for example I need longer exposure time on a tripod and has to angle it to avoid shotting its legs.. I shot angled. Hope that makes sense.
I do it exactly, but in photoscan I get many "mountains" in the objects. What is the error? :(:(
When you get glitchy and noisy surfaces which should be flat it usually means that image quality wasnt good enough for reconstruction. It usually happens when you shot with mobile. If you didnt you need to increase image quality - try to shot only with ISO 100, use tripod to stabilise your camera, capture in decent lighting conditions to make sure you have enough light to capture all nevessarry data. Bare in mind that photogrammetry software sees more than your screen can present, this is why I recommend shooting in RAW and always with as low ISO as possible. I think photoscan simply sees noise and reconstruct it into 3D mesh. The other idea I have is that surface is reflective and camera gets different data cameras angles is changing due to additional reflected information (another kind of noise). Hope that helps, if not, feel free to ping me again :). Cheers!
@@GrzegorzBaranArt thankssssssss you!!!!!! ^^
What camera did you use to do ?
Hey, I made entire video about equipment I use for photogrammetry: ua-cam.com/video/fapgOO39Dh8/v-deo.html
Regarding the camera, it was Canon80D but with some practice any DSLR camera should do the job.
@@GrzegorzBaranArt thank you !!!
I follow the same steps but the clone tool does nothing.
I believe that you are trying to use the Clone Tool on a wrong layer.
Great tutorial, but terrible appearance and disappearance of text in the video. Pixelization is not the best choice in this case.
Thanks, I agree.. this is why I dont do this anymore in my recent videos :). Appreciate feedback tho. Please let me know if recent works for you better.. I went into voiceover and subtitles. Cheers!
If you're going to use substance designer to make these kinds of textures tile, you might as well just skip the photogrammetry part and use procedural generation in SD. Photogrammetry only really shines when we capture organic surfaces that are hard to model (either by hand with sculpting or procedurally in SD).
Thanks mate but I dont agree at all.There are many ways to do textures. Some work better for one type of surface some don't. Designer would do a great job for stylised textures for sure.. trust me, I also do textures proceduraly in Substance Designer as well as sculpting them in ZBrush even using photogrammetry based alphas, I made textures using used photometric stereo or I used to use Crazy Bump or B2M many times. I also used to make textures in simple Photoshop etc as I do textures since over 15 years :D
If you aim into photorealistic high quality material, the best way to make it is to capture reality. Doesnt matter if you use photogrammetry, photometric stereo, B2M etc.. There is no easy way to achieve that level of complexity proceduraly yet. You can be close but ... just close and to be there you need a lot of time and effort. Also substance files with that level of information complexity are usually very heavy.
With photogrammetry I can make a few textures on top possible quality level just within one day. To achieve similar level of quality proceduraly in Designer I would need days per one.
Of course I am not telling that Substance sucks. From my experience I can tell it is a great tool for almost every generic type of surface, for those which don't need deeper level of surface microstory. I would say for metal, painted walls, some types of generic decals Designer is much better to captured data. Also it works great when you need to have masks under control when used as input data for more complex shaders etc.
I agree that it is easier to modify substance than scans but.. for complex substances it is not as simple anymore.
Substance help to keep everything consistent but when you capture physically correct data you also get consistency.. and finally.. scan is just a data which can be modified, you can mixed scanned data with procedurally generated or with other scans. This is why photogrammetry is so useful as it can contain physically correct surface information.. from height to color.
About organic surfaces you mentioned.. for me almost every ground surface is very complex and organic, just take a look how complex these surfaces really are. Please take a closer look on shot described as 'Source - surface details' - this is material captured for this tutorial:
www.artstation.com/artwork/rgYza
see that vegetation between each
Even if you don't see these things from the distance, it doesnt mean it is not there. These are details which makes the surface real. That noise on microlevel is very important and tells its own story, it is never random.
So if you want to create that kind of surface in Substance Designer you need to study it first, understand and try to mimic the reality. With photogrammetry I skip this part :).
But of course it is just my opinion and my experience. But I really know just a few material wizards like Joshua Lynch or Daniel Thiger who can achieve pretty close level of complexity.. but they still need time to do that.
If you simplify the material and use it next to a good one, material will always feel a bit off, even if it is hard to tell why.
Hope that helps :)
Cheers!
Weirdest way to write Million I've ever seen.