I used to spend way to much time doing the 'hard' stuff. But after enough hard surface stuff you are just able to predict how many verts you should use for shapes, if you can get away with just a few or if you need to just brute force it for best results. I am not super-pro-man at it yet, but I am learning how to predict pretty well, so I don't have to spend a crazy amount of time reworking areas. Most of the time I can just one shot complicated areas with my best guess of topology.
When I can't one shot it, I try to figure out what I did wrong in the first place and remedy that the next time I encounter that shape.
I still have a ton to learn, but feel I have come a long way since I started even this mech.
With all of that said, because I have no time limit, I am taking my time with shapes and details that I probably would just rush through in a production environment.
I used to spend way to much time doing the 'hard' stuff. But after enough hard surface stuff you are just able to predict how many verts you should use for shapes, if you can get away with just a few or if you need to just brute force it for best results.
yeah I know what you mean, theres no magic technique to doing this stuff, I guess its just hard work and practice that gives you an eye for edgeloops and poly flow that makes it faster and easier.
Here is my first test bake for the low poly. I am trying to match the high poly as close as possible. So I still have to go in and added a few edges here and there. There are also some minor errors here and there that I just haven't cleaned up yet.
Mistry - Well that is a long answer heh. I have pretty grandiose plans for this. I want in the end, to make a web based scene, where you can run around the mech in Soweto, Johannesburg. Unity+skyshop is the most logical answer at this point, but Unreal is about to head to the browsers also and it would give me a chance to learn U4's blueprints.
In the end, the static scene will be dynamic, fully realized in Soweto. I want to convey tons of movement and dynamics, because that is what the scenes involving the mech are known for.
I have not done any renders. All of these screens are real time.
Most of the screens are just 3dsmax viewport grabs using a 3-Point light setup and their nitrous viewport mode.
I used Marmoset Toolbag2 for some of the screenshots, the assault rifle and the arm+guns shot for example. That last screenshot was a realtime grab from Toolbag2 also. I just find a good cubemap, add in a point light or two to accentuate forms and get a screenshot from there.
Unoptimized, low poly first pass done! I am aiming for cinematic quality, so I have details modeled in that we would have left out for last gen. I will unwrap, bake, and then re-analyze and re-model areas to make them smoother or optimize. Any gaps will be filled with modular wires and cables that I am building out.
Substance Designer Bakes viewed in Unreal Engine 4
4k map for body
2k map for the face
4k map for all the weapons + cables (not shown)
There are plenty of bake errors due to ray intersection. Those will be fixed on my next pass.
To do:
Clone the obvious missing parts
Explode bake or bake by ID's to remove intersection errors
Add some chamfers and support edges here and there to clean up the bake
Looks very clean! Planning on reducing the map resolution by half at the end?
edit* - Oh, "cinematic quality". Guessing that's a no, then. Answers my question about why the res is so high though.
Yea I want this to be real time cinematic quality. I have no doubt that hero objects of this size will have similar budgets with this next gen, much less if they use layered textures and tiling.
At the end of the day its for the portfolio, so I set a certain visual quality bar for myself and have had to adjust to that, for example, I had to remove the head from the body uv's to give it, it's own, much higher, texel density.
Here they are currently.
Some mistakes and wasted space in there, but after packing it for so many hours I am ready to move on. Nice thing about Substance Designer is that if I wish to revist them and eek out another 5-10% of texel density later I can and it will auto update my work.
hay dude awesome work. Did you position all the uvs yourself by hand without any packing tools? I have a model with a similar amount of uv islands and it was abit of a mindf**k. Just wondering how you did it.
hay dude awesome work. Did you position all the uvs yourself by hand without any packing tools? I have a model with a similar amount of uv islands and it was abit of a mindf**k. Just wondering how you did it.
For me, as i mainly do hard surface stuff which results in a large amount of UV islands, i just unwrap portions of model manually a bit at a time. Generally i will have done the full high poly model and as i do the low poly model i unwrap as i go, then gradually attach sections of the model to each other, scale appropriately so pixel density is consistent, then when everything is attached i start manoeuvring into the atlas square or rectangle.
I make use of the auto flatten mapping tools in 3D Max but only do a piece at time, as if you do it for large portions of a model your UV shells will be illogically positioned making texturing frustrating.
WOuld like to hear Quacks workflow though also....
Wait SD auto updates if you change your UVs? Say wha?! Thanks so much for the UVs though!
As long as you hadn't painted custom masks, the changes will auto-propagate. That said, I will have to use custom masks, so when I get to that point I will have to be sure the uv's aren't changing if I want to avoid doing them over.
hay dude awesome work. Did you position all the uvs yourself by hand without any packing tools? I have a model with a similar amount of uv islands and it was abit of a mindf**k. Just wondering how you did it.
Yea these were all hand placed. Auto-pack algorithms are great for doing quick packs to test bake, but I can very easily double/quadruple the texel density by handpacking over using auto-pack. I spent over 10 hours on UV'ing alone. I use the standard max tools + polyunwrapper. I would prefer to use Headus UVLayout, but it is too expensive.
WOuld like to hear Quacks workflow though also....
Whisky
I just planar map large areas that I know either need to be a unique uv shell or a seperate smoothing group. I relax / straighten and move on. I don't normalize shells until all the shells are unwrapped. I straighten as many uv shells as possible to make them rectangles, as far as distortion will allow. Then I go through and a straighten each island along an edge to make it horizontal or vertical so they bake better. I pick player faces UV islands and scale them up 125-175%, depending on the need. I scale down non player facing shells to 50-75% or just remove them from the mesh. Start with the big shells and pack those, then move to medium, rinse and repeat. My initial UV pack is not the final pack. I pack once, then scale to the UV extents and then re-jigger the UV shells to get more texel density. If the model doesn't have 500 uv islands I will re-pack again if it will get me at least 10% more texel density. I usually stop at the 3rd repack. All of this is using hotkeys only, click as few buttons as possible. Nothing mind blowing, just tedious work.
I ended up going with the standard High-->Low-->Explode-->DuplicateLowAsCage-->Xnormal-->Toolbag2
Thanks for the answer, I'm just curious about how people bake such a complex mesh. I tried "match only material ID" in max and it took forever to render an AO map. So, I guess explode is tedious but in the end of the day it does the job
anyway, waiting for your texture, I bet it gonna be great
Beef, EQ, Mistry, and Swall - Thanks! SHould have preliminary colors and base textures on tonight!
MeshPotato - I plan on doing as much in Designer as allows. I wouldn't mind going into Painter and adding custom details, so yea that will probably happen.
Here is something I wasn't plan on doing, I was asked by a co worker what is the best rendering software and I got laughed at when I suggested that Toolbag2 was all you need. So I whipped up a quick GI render test to see if I can get a nice looking clay render that takes no render time. This was fairly quick and my first iteration and could be improved upon drastically:
Now, if I may ask: How long did this piece of sub-d nightmare take you?
If done in off-hours how long should it take to a professional? Like say, for a movie or something?
I'll keep an eye on the thread, good job!
Thanks!
I don't really know how many hours I took on the high poly version. I would guess over 100.
As for time it would take in a production setting, that will vary wildly. If you were going for next gen quality I would say 3 to 4, 40 hour work weeks for the high/low/bake would be sufficient at this quality. Less if you were targeting a smaller low poly tri count.
VERY WIP progress of my Substance Designer first try.
The Mech was black in the movie, but was White in the concepts and was MUCH cooler to me as white, so I will be going for the Mostly white color scheme.
Still to do:
-Specific wear, bullet impacts, isolated gouges.
-Give more logic to the wear and tear/ remove cavity mask from AO collected dirt
-Iterate on the alien chrome material once more.
-Add a top layer of light colored dust/dirt, masked by gradient from ground
-Add Alien markings and incorporate by making them look painted on with a bit of normal map relief
-Normalize color values and materials in Toolbag2
-Add cable details
For some reason, I'm noticing that most SD texture work tend to look noisy and weird.
Is it perhaps because of the Curv/AO maps that SD produces?
Don't get me wrong. I adore SD, and use it daily, but it seems like the mesh-adaptive generators tend to be abit finicky. In the end, I tend to hand paint alot of detail instead.
Replies
On to the Arc Gun:
Update:
Thanks Sam!
Finalized high poly on the arc gun. Going to take the two guns and an arm into Toolbag2 tonight for some sexy screenshots.
I used to spend way to much time doing the 'hard' stuff. But after enough hard surface stuff you are just able to predict how many verts you should use for shapes, if you can get away with just a few or if you need to just brute force it for best results. I am not super-pro-man at it yet, but I am learning how to predict pretty well, so I don't have to spend a crazy amount of time reworking areas. Most of the time I can just one shot complicated areas with my best guess of topology.
When I can't one shot it, I try to figure out what I did wrong in the first place and remedy that the next time I encounter that shape.
I still have a ton to learn, but feel I have come a long way since I started even this mech.
With all of that said, because I have no time limit, I am taking my time with shapes and details that I probably would just rush through in a production environment.
Wires:
The arm + the assault rifle + the arc gun:
yeah I know what you mean, theres no magic technique to doing this stuff, I guess its just hard work and practice that gives you an eye for edgeloops and poly flow that makes it faster and easier.
Ged - Practice is nice an all, but wanting to know how to do things better is key.
It will be about 8.5k tris for each arm.
What are you plans for your final presentation (money shot) ?
Mistry - Well that is a long answer heh. I have pretty grandiose plans for this. I want in the end, to make a web based scene, where you can run around the mech in Soweto, Johannesburg. Unity+skyshop is the most logical answer at this point, but Unreal is about to head to the browsers also and it would give me a chance to learn U4's blueprints.
In the end, the static scene will be dynamic, fully realized in Soweto. I want to convey tons of movement and dynamics, because that is what the scenes involving the mech are known for.
I have not done any renders. All of these screens are real time.
Most of the screens are just 3dsmax viewport grabs using a 3-Point light setup and their nitrous viewport mode.
I used Marmoset Toolbag2 for some of the screenshots, the assault rifle and the arm+guns shot for example. That last screenshot was a realtime grab from Toolbag2 also. I just find a good cubemap, add in a point light or two to accentuate forms and get a screenshot from there.
Am lookin forward to more
Unoptimized, low poly first pass done! I am aiming for cinematic quality, so I have details modeled in that we would have left out for last gen. I will unwrap, bake, and then re-analyze and re-model areas to make them smoother or optimize. Any gaps will be filled with modular wires and cables that I am building out.
70k Tris.
Thanks beef!
Thanks!
Latest Update
Substance Designer Bakes viewed in Unreal Engine 4
4k map for body
2k map for the face
4k map for all the weapons + cables (not shown)
There are plenty of bake errors due to ray intersection. Those will be fixed on my next pass.
To do:
Clone the obvious missing parts
Explode bake or bake by ID's to remove intersection errors
Add some chamfers and support edges here and there to clean up the bake
edit* - Oh, "cinematic quality". Guessing that's a no, then. Answers my question about why the res is so high though.
Yea I want this to be real time cinematic quality. I have no doubt that hero objects of this size will have similar budgets with this next gen, much less if they use layered textures and tiling.
At the end of the day its for the portfolio, so I set a certain visual quality bar for myself and have had to adjust to that, for example, I had to remove the head from the body uv's to give it, it's own, much higher, texel density.
@Dan
Here they are currently.
Some mistakes and wasted space in there, but after packing it for so many hours I am ready to move on. Nice thing about Substance Designer is that if I wish to revist them and eek out another 5-10% of texel density later I can and it will auto update my work.
For me, as i mainly do hard surface stuff which results in a large amount of UV islands, i just unwrap portions of model manually a bit at a time. Generally i will have done the full high poly model and as i do the low poly model i unwrap as i go, then gradually attach sections of the model to each other, scale appropriately so pixel density is consistent, then when everything is attached i start manoeuvring into the atlas square or rectangle.
I make use of the auto flatten mapping tools in 3D Max but only do a piece at time, as if you do it for large portions of a model your UV shells will be illogically positioned making texturing frustrating.
WOuld like to hear Quacks workflow though also....
Whisky
Thanks, I am excited too!
As long as you hadn't painted custom masks, the changes will auto-propagate. That said, I will have to use custom masks, so when I get to that point I will have to be sure the uv's aren't changing if I want to avoid doing them over.
Yea these were all hand placed. Auto-pack algorithms are great for doing quick packs to test bake, but I can very easily double/quadruple the texel density by handpacking over using auto-pack. I spent over 10 hours on UV'ing alone. I use the standard max tools + polyunwrapper. I would prefer to use Headus UVLayout, but it is too expensive.
I just planar map large areas that I know either need to be a unique uv shell or a seperate smoothing group. I relax / straighten and move on. I don't normalize shells until all the shells are unwrapped. I straighten as many uv shells as possible to make them rectangles, as far as distortion will allow. Then I go through and a straighten each island along an edge to make it horizontal or vertical so they bake better. I pick player faces UV islands and scale them up 125-175%, depending on the need. I scale down non player facing shells to 50-75% or just remove them from the mesh. Start with the big shells and pack those, then move to medium, rinse and repeat. My initial UV pack is not the final pack. I pack once, then scale to the UV extents and then re-jigger the UV shells to get more texel density. If the model doesn't have 500 uv islands I will re-pack again if it will get me at least 10% more texel density. I usually stop at the 3rd repack. All of this is using hotkeys only, click as few buttons as possible. Nothing mind blowing, just tedious work.
There are still plenty of errors in the bake because of the quick Push to the cage I did, will clean up tomorrow.
55k Tris, without the head / guns. Will end up in the 80k range.
Awesome work mister.
Thank you!
I ended up going with the standard High-->Low-->Explode-->DuplicateLowAsCage-->Xnormal-->Toolbag2
Can't wait to see the final textured version, I'm sure you will blow everyone's socks off.
Thanks for the answer, I'm just curious about how people bake such a complex mesh. I tried "match only material ID" in max and it took forever to render an AO map. So, I guess explode is tedious but in the end of the day it does the job
anyway, waiting for your texture, I bet it gonna be great
Latest progress, bake cleaned up enough. Now onto color blockouts and Substance Designer
65k tris, 1 4096 map for the body, 1 2048 for the head. Toolbag2.
Can't wait for the textures, Gonna follow it closely.
Will you use Substance Painter for hand painted detail? Or Designer only?
MeshPotato - I plan on doing as much in Designer as allows. I wouldn't mind going into Painter and adding custom details, so yea that will probably happen.
Here is something I wasn't plan on doing, I was asked by a co worker what is the best rendering software and I got laughed at when I suggested that Toolbag2 was all you need. So I whipped up a quick GI render test to see if I can get a nice looking clay render that takes no render time. This was fairly quick and my first iteration and could be improved upon drastically:
Ok, ok, enough playing, back to texture work!
Now, if I may ask: How long did this piece of sub-d nightmare take you?
If done in off-hours how long should it take to a professional? Like say, for a movie or something?
I'll keep an eye on the thread, good job!
I don't really know how many hours I took on the high poly version. I would guess over 100.
As for time it would take in a production setting, that will vary wildly. If you were going for next gen quality I would say 3 to 4, 40 hour work weeks for the high/low/bake would be sufficient at this quality. Less if you were targeting a smaller low poly tri count.
VERY WIP progress of my Substance Designer first try.
The Mech was black in the movie, but was White in the concepts and was MUCH cooler to me as white, so I will be going for the Mostly white color scheme.
Still to do:
-Specific wear, bullet impacts, isolated gouges.
-Give more logic to the wear and tear/ remove cavity mask from AO collected dirt
-Iterate on the alien chrome material once more.
-Add a top layer of light colored dust/dirt, masked by gradient from ground
-Add Alien markings and incorporate by making them look painted on with a bit of normal map relief
-Normalize color values and materials in Toolbag2
-Add cable details
Is it perhaps because of the Curv/AO maps that SD produces?
Don't get me wrong. I adore SD, and use it daily, but it seems like the mesh-adaptive generators tend to be abit finicky. In the end, I tend to hand paint alot of detail instead.