'Nostromo' environment study

[NiZu]

Hi,
This is a personal project i started to test, improve and solve some issues related to the techniques and workflow i’ve used in modelling and shading the assets made for Mango , main question was (and generally is , for all sorts of environments) making lots of models and textures , make them quickly, detailed and not too slow to render

I looked for a good test case, and watching again the opening scenes of ‘Alien’ with the camera flying through the ‘Nostromo’ seemed perfect : great design by Ron Cobb for inspiration , reasonable to work on 2-4 quick shots of interiors, recycling some assets and building some new.

This is still work in progress ! shot 1 is basically done , 2 needs a few detail models more , 1-2 more shots to add to test other cases.

For texturing i used the tecniques described in the the Mango env art videos (http://www.youtube.com/channel/UCDSBZIemT12ZTjTqwTFjd6Q)
There’s no custom painted textures here, all tileables and dirtmaps , and only a few are baked AO, the much lighter ‘dirt vertex color’ proved to give some great results in many cases .

The key point was to test more on the renderspeed , so have these scenes animated : while using dirtmaps and blending can produce some great materials the question remains on how heavy that is compared to using 1 higher res. texture per channel…
There’s no quick universal answer of course , but in these tests i managed to get rendertimes per frame of 2-3 minutes (full HD , on a i7 3930k cpu) … with direct light (0/0 dif/glossy bounces) and AO , using non-progressive mode.

The question with complex materials (texture wise) in Cycles seem to be that each ‘sample’ means recalculating the whole node tree , so it’s quite slow when using 100s or 1000s of samples , but it’s dead fast when using non-progressive with low AA samples. (these shots use 6-8 AA samples)

However, non progressive is tricky to setup , and right now it only seems to be convenient when using direct light only or vey limited GI.
I made a tutorial on the subject : http://www.youtube.com/watch?v=7Hb5fJ00P2w

So … interestingly, i ended up using Cycles to do some quite ‘old-school’ setups : using 0 glossy bounces means basically no raytraced reflections , only speculars … but still often some sweet looking area-light speculars
And there’s no GI in these renders , only AO … but when trying to keep renders very fast even Vray won’t guarantee an easy life (tons of parameters to tweak, blotches and other complicate glitches…) while AO is easy and reliable …

So i’m testing and figuring out just how much can be obtained with these setups . For sure they have some serious limits when trying to achieve photorealism , but for animation or architecture videos … fast and reliable is important , and the question remains on how limiting it is for various situations (soft shadows are slow anyway, chromes and mirrors require 1 bounce of glossy, etc… )

But enough technical render stuff (Cycles … you got to love it, but as usual with render engines and lighting you’re never done testing and learning workflows :)) below is a frame with some passes breakdown :

Thanks for comment and critique , i’ll post more about other aspects (like compositing) in updates.

[JoolsMcFly]

Very, very good job! I’ve subscribed to your channel and will hopefully watch your vids sometime in a not so distant future.

[gundamf91]

wow impressive!

[ctdabomb]

o_O that’s pretty cool!

[Guiseppe]

Hello NiZu nice result, but how do you achieve Specular in Cycles?

[ng-material]

very nice,also bookmarked your youtube videos.

[NiZu]

Thanks everyone!

Guiseppe : ‘speculars’ can’t be turned on or off in cycles, if you have 0 glossy bounces the only kind of reflection you get is lamps directly visible by the surface , that’s the most common definition of ‘specular’ …

The classic definition of specular also implies the lamp is seen as an infinitely small point … in Cycles ‘0 bounces glossy’ do take into account lamps size , and that makes lot of quality difference , but they’re still quite fast and require just a few samples.

Also, check ‘direct light’ preset in Cycles : it uses 1 glossy bounce , because ‘bounce #1’ isn’t properly GI, it’s raytraced reflections of the scene on other objects … and it makes sense to consider bounce #1 of glossy as necessary for ‘direct light’ render.

This sounds confusing … but it’s actually much more logical than older engines with all their different system per each part of the render, duplicate features etc…

So it’s a bit of a personal idea to call them ‘speculars’ , i got the idea from Vray having a similar feature called ‘area light speculars’ to improve old-school point-light specs.

A technical note … i guess old style ‘speculars’ were not raytraced calculations … everything in cycles is raytraced … but still , imho the name fits.

edit : so what i did is just set glossy bounces to 0 , and got ‘speculars only’ render … of course it has big limits : no mirrors are possible and pure chrome materials will look wrong (quite dark)
I don’t know yet of a way to do that only for some materials (maybe : light path nodes to limit some materials to spec only …)

[comeinandburn]

really impressive results! I’ll be following your progress:)

[Mrdodobird (Ian Hubert)]

Nicolo!!! I just watched that video 6 times! Loving all the detail/lighting/texturing! And those chairs and that desk are crazy rad design wise. The back of those chairs! Is that your own design? Phenomenal. Looking forward to seeing what else you do with this!!!

[NiZu]

Hey, thanks!
Since it’s you mentioning it , i have to say there’s definitely something of the design of the chair you modelled for that early Mango pic …i think i got from there the idea that barber-style chairs can be awesome sci-fi props

I did not use a precise ref for the env. but tons of partial refs for everything !
In general, i looked at Ron Cobb’s style : Alien stills , but also sketches : http://propsummit.com/upload/585/airlockrcobb.jpg(for the corridor section and doors)
And other cg ‘remakes’ , this one in particular : http://features.cgsociety.org/newgallerycrits/g40/422540/422540_1323953738_large.jpg
(actually the point of the 2nd shot was to make a ‘clean’ env. that would look realisticly worn but not ruined … still working on that. )

Or, small stuff : the ‘caps’ on the back of the chair are copied from here : http://cghub.com/images/view/315304/
On the knee of the robot , there’s a black round cap that looks like a plain, everyday plastic cap or part of a bottle …trivial but that’s realism !
Only problem is the time and precision it takes to be that accurate i remodelled that twice …and it’s still not 100% ok in proportions, bevels and shape

Reference hunting is just great … often there’s no time to exagerate with this stuff , and i normally cut a lot of corners modelling, but once in a while it’s good for learning to really insist on accuracy

[NiZu]

Quick update :

For comparison , here’s a frame rendered with 1diffuse,1glossy bounces … adding little apart from noise , that’s something to keep in mind : sometimes GI bounces contribute a lot less than you expect to the final image…

Of course i’m not generalizing : if GI was free in rendertime, this same scene with just a different lamps setup (less, stronger lamps) could use GI , and definitely add to realism and atmosphere.

For glossies , only the table is showing a big difference , and about that … i tried the light path node to control number of bounces per material : problem is , you control the material as “seen” by other objects , what i think i’d need is to control the opposite : per each material decide how many bounces will be cast from that point to objects around …

The render took 35 mins. (and it’s not acceptable for animation) at 300 samples , using progressive (i tested and, at 1dif-1glos bounces N-P was noisier at same rendertime).
For comparison the 0-0 bounces render took 3 mins ! … and again, you can’t make any general rule from this (0 bounces or 1 , direct-no-glossy or GI render … can’t be compared technically) but it’s the kind of things you have to be aware (with Cycles more than other engines, imho)

more soon ,

[NiZu]

Hi,
Update , i put shot 1 and 2 toghether , worked a bit on the smoke sim, and tested a new variant of the 3 steps blend - nodegroup:
You can see it in the main pieces of the corridor (i.e. the orange metal plates): similar to the other one, it uses a dirtmap to create areas of dirt/base-paint/scratches , except instead of blending smoothly it uses a sharp threshold, it’s good for ‘chipped paint’ rather than ‘worn out’ surfaces.

Putting togheter the meshes for the 2 shots caused some trouble , but that’s exactly what i wanted to test here.

The env. pieces are modelled in different scenes , i.e. : 1 for the corridor , 1 for the “hub-room”: these ‘rooms’ then go into an assembly scene as dupligroups.
That makes editing meshes easier when you have many objects.
And makes it easier to i.e. make another scene for smoke, just add those 2 dupli-objects and you have all the meshes linked and always updated, then you can add new lights , override materials etc… all you need for a smoke pass.

Great, but has some risks : you must not mess up the hierarchy of these links , the rules are never to link backwards and try to avoid linking sideways, ie. here i had issues with the doors , a door panel is a single piece that gets grouped in a bigger group (a complete, animated, door) and the door is part of the ‘room’ group …
No backward linking, ok. But then you use the ‘panel’ as a non-animated piece of another room … so now that mesh is instanced as part of ‘door’, inside ‘roomA’ and directly inside ‘roomB’ … already getting complicate and potentially messy.

The issue you might get in these cases is meshes appearing out of place in the final assembly , mostly it’s a bug if Blender fails to ‘solve’ the hierarchy between objects , but it’s also easy to create totally unsolvable situations.
In any case it’s also easy for the user to think a bit and simplify the ‘tree’ of links and groups, and make everything solid and stable.

(Btw , in Mango i must have been more careful and never had issues with meshes , but i had a similar case with material node groups, … it’s an important skill understanding hierarchy , links and grouping and knowing when and how complicate you’re making it)

That’s the downside , but besides being useful for organizing modelling , it seems grouping also affect rendertime by affecting the bvh build time :
This frame rendered in 3 mins with 1 min bvh build time or 5mins (3 mins bvh) with a different grouping !

Consider the main room, it’s an octagon , many pieces repeat per 8 slices , i could just instance (alt+d) every piece 8 times , or make a group for the ‘slice’ and make 8 duplis … the latter is much faster to build bvh !

I tried with a simple test scene , and i can’t find a simple rule in this , but definitely : not all instancing methods are the same for bvh , having dupligroups , external links with more or less sub-grouping affects speed a lot !

[ferrettank]

Wow this is amazing!

[Dacoolinus]

This looks fantastic!!!

[DE3D]

Very nice job; Alien is one of my all-time favorite movies (together with Blade Runner and Prometheus - Yes I have a big man-crush on Ridley Scott! ) I’d love to see a shot-for-shot remake of the opening (from the end of the titles to the moment the crew awakes) in Cycles… You’ve definitely got the feel of the Nostromo down pat!

Keep it up; I’ll be following this one closely.

[NiZu]

Thanks!
DE3D: Indeed matching the opening sequence as close as possible would be a great challenge , even better for learning ! but of course it’s great fun to take some liberties in design ( and also i plainly wanted to cut and simplify the workload

Next piece is also based on Alien(s) : my take on the famous apc (Ron Cob’s design again) it’s a bit of that , with some elements of an hum-vee. (plus of course browsing 100s of concepts on the web to steal some inspiration ! )

I’m planning to use this in the 4th and last shot : again this project is mainly an attempt to “chain” toghether many small, personal learning tests into something bigger , so i’ll keep working in my free time (possibly for the rest of the year…)

Main point for this was practice texturing on some kind of military vehicle.

Below a sort of texturing ‘breakdown’ , it’s my usual dirtmap and stencils workflow, but in a slower, refined variant : instead of quick vertex colors for base paint (or even for dirtmap) it uses various layers of bitmaps : painted color , dirtmap from batchbake with painted tweaks, painted mask for mudsplatters. (painted meaning Blender texture painting) plus photo based tileables (blended by masks or overlayed).

I’ve made a 3rd variation of the stencil nodegroup with a slider for ‘contrast’ : it blends between smooth and sharp transition for the dirtmap. This way scratches/dirt/rust patches have a precise, sharp shape but also a bit of falloff.

I’ll try to record a video going over the process : the ideas/techniques are the same of previous videos , but it’s practice and the eye for real life materials that make the difference in these things (that’s why i keep practicing …and i did maany passes of corrections on this one ! )

I’ll also put togheter a “resources” page on my blog with the tutorials and some files (new nodegroups downloadable soon)

Attachments

[Jake_Shadow]

Awesome job with the animation! Rendering animations with Cycles sounds pretty scary, usually I’ll stick to the internal for that. But your outcome looks very impressive.

[Selrond]

Well!
Even though this is in “Focused critique” thread, I had to skip those several “helpful” comments like “Awesome job” and such, which’s kinda bad and lowers the level of profesionality…
Anyways:
I’ll begin with good things:
The amount of details and texturing is OUTSTANDING man! How much time did you spend on them? better don’t wanna know…
On the other hand though, the animation ant the first scenes you posted were like an absolut chaos for me… Everything reflected everything and such… I’d add more focus to a particular things… I didn’t know where to look at first… my eyes were like mad trying to catch all the details and it distracted me a lot…
So just one thing: Bring on the focus on something particular to force the viewer’s look and attention where you want it

[JDaniels (DanBrownCGI)]

The materials on these items are top notch. I’ve been at a loss to say what I think should be improved upon. Though I’m not a fan of running multiple passes, sometimes it’s necessary, and you’ve got it down to a science. Of course, what Selrond said makes sense from a strictly compositional sense, but I’m sure there will be some pretty cool stuff to focus our attention on in short order, and for now the environment deserves a chance to stand in the spotlight. =D

Great work!

Dan

[NiZu]

Jake Shadow : Thanks, i guess everyone has those same worries about cycles and animation, and indeed that’s the point of this test.
But also the question is : Cycles is slower and has more features than BI , so … can we use it more alike BI (…that is no GI) ? and get faster renders when needed ?
So far results have been positive and interesting , with the only big question of not being able to use ‘true’ raytraced reflections (…1 bounce i mean) only for some materials , and ‘area light speculars’ for the rest…

Selrond : Thanks , good point , i get what you mean about focus .
i did care about something related trying to use different light hues and shadow gradients to have ‘big shapes’ to help readability and avoid -as i tend to do- to fill the scene with tons of equally small/medium details .
But indeed there is a lack of focal points (…the fact that there’s no characters doesn’t matter : objects can be focal points) … i’ll keep that in mind !

JDaniels : Thanks ! i’m a big fan of your Giant Spaceship project

By multiple passes you mean the separate renderpass for the smoke ?
That’s always clunky and messy … needing multiple renders and then putting stuff toghether , keeping stuff linked and updated … but i needed practice for when it’s necessary.
Same would be for matte painting … i don’t love the idea , but i’d like to explore and learn the workflow for when you really have too many polies for a single pass…