Demystify Archviz in Houdini

[neonbulbs80]

Got an idea some months ago on how most people use Houdini mostly for Vfx. In a way I always think that CG softwares are just a tool, the difference is how can we use it to get what we want. So I'm trying to demystify the usage of Houdini to do architectural visualizations. My goal in here is not for trying to proof something it's just an exercise for me to get use to with Houdini, since I'm still quite new with Houdini and I found it fascinating.

Ranging from procedural modelling, digital assets, lighting and rendering, this personal project is my hobby other than my demanding deadline at my workplace. So, I hope be able to finish it and hope you guys can give suggestions and critiques.

Cheers..

[Tom]

At this moment it looks good, I hope to see that shelf scene textured

When I tested Houdini to render studio type scene, it actualy did a better lightning than Vray, at least it looked the way I want.

Cheers,

Tom

[dynamoanders]

Nice ones Donny!

I do agree that Houdini is way more powerful than "just" doing effects.

Personally I have been looking to find a procedual way to do 3D ever since I started, and thought for a long time that Maya was the way for me.

Now that I have found Houdini, I am just so happy, and walk around thinking how to solve problems in it everyday!

I especially like the way you can create really nice procedual shaders, displacements and renderings and I think that´s where Houdini is really underestimated so far (like architectual renderings etc).

Anyway, keep up the good work!

[Fabiano Berlim]

That's a great idea.

Indeed 3D software is just a tool, but the tool can also influence the work.

Houdini easily give you the math, expression and procedural way of thinking without loose the destructible way of thinking.

That can help your work step out. Can only help, the rest is up to you.

Good art always come with creation constrains, such as: Style, philosophy or even the tool.

I know it can sound odd, but Houdini can give you others creations constrains.

[renderfan]

Houdini is great tool to model procedural geometry but for archviz Mantra can not compete with Vray or Mental Ray. It is much slower and has not good quality! I am not very experienced with Mantra and I do not want to blame Houdini but everytime I tried to render I was disapointed. Vray can render same in 1/10 of time.

[neonbulbs80]

@Tom, @dynamoanders, @Fabiano Berlim : Thanks for the comments. I'll be hoping to get started on the texturing side soon and start to do the final lighting.

renderfan :

Houdini is great tool to model procedural geometry but for archviz Mantra can not compete with Vray or Mental Ray. It is much slower and has not good quality! I am not very experienced with Mantra and I do not want to blame Houdini but everytime I tried to render I was disapointed. Vray can render same in 1/10 of time.

Thanks renderfan. Yes, I do agree with you. But the reason I do this because I happen to like Houdini as software of choice right now for maybe the rest of my work. Been using other softwares for Archviz like VRay for Maya, Mental Ray in Softimage and Modo in productions. The thing is, it's a challenge for me to do things in a different way. I know how to light archviz using VRay and MR but I found out that you can't treat Mantra to behave the same thing as those renderers. Surely what I do right now got cheats and caveats to get the renders done and one thing that I learn that Mantra is just a different beast to handle and that's why I submit WIP in here so that if I got problems with Mantra hopefully other people who got more experience with it here in OdForce can help me to figure it out .

Anyway just add up my last piece of object at the last top shelve. Playing around with metaballs and polywire sure do comes in handy. Now I guess on to shading!

Cheers

[Sifis]

Got an idea some months ago on how most people use Houdini mostly for Vfx. In a way I always think that CG softwares are just a tool, the difference is how can we use it to get what we want. So I'm trying to demystify the usage of Houdini to do architectural visualizations. My goal in here is not for trying to proof something it's just an exercise for me to get use to with Houdini, since I'm still quite new with Houdini and I found it fascinating.

Nice Idea neonbulbs80, I absolutely feel the same. Houdini can do much more than just vfx and mantra is getting better and better each version. Unfortunately I still think it's much slower than say Vray but given the fact that mantra is much more powerful and flexible than Vray you can find many ways to optimize it's behavior by writing your own optimized shaders or developing a more efficient pipeline to render faster. Overall though, despite the slowness of mantra I think it's the most flexible and complete renderer around. It can render everything out of the box, from hard surfaced architecturals and cars with instant path traced GI to volumes with stochastic sampling (really fast), point rendering/instancing, rendertime subdivision, procedurals and displacement with no extra cost in rendertime. And all that with fast motion blur and DOF. Anyone who ever tried to add all those things in one render with other renderers knows what I 'm talking about. So yes, I think that although mantra has a lot of room for improvement (speed wise), I think now is quite mature for archviz or product rendering. Keep it up!

[renderfan]

I agree, Mantra is very flexible and you can do everything. And I love to render with Mantra but for archviz and GI I think it is not usable. The advantage for DOF and Motion Blur is no longer true, because Mental Ray have Unified Sampling and makes render very fast. I would really like to do comparison but I do not know Mantra good enough to optimize the render, but if some Mantra-spezialists can help it would be nice. I do not want the "what is best renderer" war but would really like to see if Mantra could compete with Mental Ray or Vray for archviz.

Thank you!!

[Andz]

Hi guys,

I have struggle before, giving a try at arch lighting in Mantra. But never had the time to take my tests to anything usable.

I have THIS scene that I have done in the past and whenever I'm trying a different render engine, I use it as a benchmark, not only for speed, but for quality and ease of use as well.

The renders you see there are from vray. And so far, are the best results I have. In second came Mental Ray but I don't have the renderings anymore, but I remember that it took something like 2x or 3x to get the same indirect lighting quality. I have also tried with the demo version of fry render, maxwell and octane, all of them are PBR renderers and it took a long time to resolve to something near usable, specially the small recessed spot lights. Octane was the fastest, that used my GPU, but it couldn't handle the whole scene due to the RAM limitations from my card (only 256mb).

I'm not sure I'm allowed to offer all the objects in the scene, but I can upload most of it and maybe you guys can have something to play and see what you come up with.

I don't want to hijack this thread, so I'll contact the admins, and see if they would like to turn it into a lighting contest.

Edited March 13, 2012 by Andz

[DaJuice]

Andz, go for it. Would love to give it a crack.

[michael]

why not use one of these scenes:

http://www.3drender.com/challenges/index.htm

[jim c]

Houdini is great tool to model procedural geometry but for archviz Mantra can not compete with Vray or Mental Ray. It is much slower and has not good quality! I am not very experienced with Mantra and I do not want to blame Houdini but everytime I tried to render I was disapointed. Vray can render same in 1/10 of time.

Could you be more specific? I'm generally curious to here what makes you think mantra doesn't have the quality of MR? If you're getting 10X speed differences, then that sounds like a really misconfigured mantra output node. I have noticed that to use mantra well for rendering you have to learn *ALOT* about rendering principles, and all sorts of other gritty 3D details. Which is unfortunate, but that's the way it is.

[renderfan]

Could you be more specific? I'm generally curious to here what makes you think mantra doesn't have the quality of MR? If you're getting 10X speed differences, then that sounds like a really misconfigured mantra output node. I have noticed that to use mantra well for rendering you have to learn *ALOT* about rendering principles, and all sorts of other gritty 3D details. Which is unfortunate, but that's the way it is.

Hello Jim, please see here:

[jim c]

Ahh, OK. Well one thing I have noticed - something I tried a few weeks ago - when making a really simple test scene like this mantra kind of stunk, but when I did a different scene, something I was actually going to conceivable use, with reasonably complex geometry (over 350K polys), displacement, etc, then all of a sudden it made a significant difference, and in fact was rendering as fast or even a little faster than Modo (I don't have VRay). I don't know why this would be the case, but it does seem to make a difference. But I am *far* from being any sort of expert on it.

[neonbulbs80]

Hello Jim, please see here:

Ahh, OK. Well one thing I have noticed - something I tried a few weeks ago - when making a really simple test scene like this mantra kind of stunk, but when I did a different scene, something I was actually going to conceivable use, with reasonably complex geometry (over 350K polys), displacement, etc, then all of a sudden it made a significant difference, and in fact was rendering as fast or even a little faster than Modo (I don't have VRay). I don't know why this would be the case, but it does seem to make a difference. But I am *far* from being any sort of expert on it.

Hi renderfan, jim c

Just see the discussion between you both and this is what I think. Even though I only been using Mantra only for a little while, I been using MR for 5-6 years in production, VRay is around 2, and last year I tried on Modo too. And these last two years our studio is using 3Delight for production. Just like on my previous reply those renderers have totally different characteristics on how to render things. It's a bit hard to match up qualities from one renderer with another from straight render, that's why you have render buffers or should I say AOVs so that the compositor can work things out and the final result should have the same continuity even tough they are coming from different renderer.

Where as for speed wise for GI solutions MR and VRay are using secondary bounce maps to simulate bouncing of lights such as Final Gathering, Irradiance Maps, Light Caches and Photon Maps. These GI solutions are basically to "simulate" but my guess is they are not 100% physically correct, because you can play a lot with those numbers to achieve what kind of results that you want versus time and deadline of course. For simple test scenes just like in http://forums.odforce.net/index.php?/topic/15114-render-question/page__view__findpost__p__93561 MR and VRay of course able to produce things fast with their GI solutions where as Mantra PBR are slow and very grainy.

The first time I use Mantra and PBR, I treated them the same way as VRay and MR but I found out that it didn't work. Slow and grainy are always be results, but that doesn't mean Mantra can't produce the same thing like MR and VRay, there are other methods that I actually used to improve rendering speed and GI solutions using Mantra PBR. Since PBR is doing path tracing for it's GI method, you can't rely on one light for Mantra to do the rest of the bounce because I believe it's going to take super long and many many samples to smooth it out. My way is set MPlay to 2.2 gamma and work my lighting from then on. For the current project I'm working on my diffuse bounce is only 1, because I have enough lights to simulate the bouncing and the render times doesn't take very long. Also you can still cheat things up like which object that you want to show in reflections so that Mantra doesn't have to do additional raytrace for every single objects.

The other thing is memory management. VRay and MR GI solutions are mostly resolution dependent. You can try test rendering your test scenes in higher resolutions such as HD and check your RAM consumption when you render in low res and high res, most likely the bigger the scene is MR and VRay GI performance starts to drop vs the RAM that you have. We have a test with one of our project in the studio where we try to render a big room full of servers with VRay and the RAM consumptions for IR map and Light Cache can take up more than 8 GB of RAM, some of our machines start rendering very slow due to VRay that needs to swap virtual memory from hard drive and actual RAM in order to be able to render an HD resolutions frame. Where as 3Delight that doesn't do raytracing can render that easily in HD with less than 2 minutes and lower memory consumption with multiple passes and we can fake the GI in compositing.

So I guess that rendering wise, renderers have different approach and different consumers as well. For instance architects and designers who don't want to know technical stuffs and hassle to setup rendering can use Maxwell Render, because it's easy setup, put lights, set simulation time and leave it for the night. Where as other renderer that is need more setups generally use for productions like 3Delight, Renderman, Mantra, VRay and MR because we can tweak their values depends on quality vs time that we usualy sets in the deadline.

Sorry if it's too long to explain, but it's just my 2 cents.

Anyway already start doing the texturing, hopefully to be able to post new WIP pretty soon. Thanks for all the comments guys.

Cheers..

[SreckoM]

Great thread.

I also started something similar for same reason, here you can see current results LINK

Unfortunately I do not have much free time, and probably will not have soon, to continue working on this.

I really do not experience so drastic difference in render times, at least not in this stage.

[neonbulbs80]

Phew been a long time since I updated this thread from March.

Work has been the case. I'm still struggling to find some time to continue on the shelve project. But I think it's a bit too complex for me to understand and get used to with PBR renders.

So I just use smaller test case but still in relation with Viz projects. Modelled some simple coffee cups and light them up as a product shots. After trial and error with render times and many hours reading the help files, setting up PBR renders with Photon Mapping in an enclosed area works the fastest and giving much better results with quality and balanced with render times. Previous attempts with only using reflector and no photon mapping can be a whopping 1 hour render for 720x540 frame with depth of field. Finally with photon maps I can get away with 9 minutes render time with smoother noise and additional bounce lights on darker areas such as the small opening on the left cup that slightly opened.

I'm still struggling to know how to make smooth workflow for photon maps in animations to reduce noise and flickers. Maybe some of you guys can shed some lights in here?

Cheers..

[qbick]

its better to implement something like vray's lightcache or final render's harmonics instead of using old photon mapping.

photons has alot of disadvantages compared to lightcache.

one of them - there is no need to set up photons count - only detail treshold.

photon maps cannot properly distribute samples on detailed surfaces or corners - thats simply impossible with photon maps to distribute samples based on distance between surfaces because its light emitter-based.

i personally think that vrays' bruteforce with lightcache is the best solution of rendering scenes where flickerless highquality GI is needed.

mantra has a similar approach - photonmaps with bruteforce detail enhancement.

but photonmaps should be replaced, imho.

for now photonmaps can be used only with still images. well, it can be used in animations to but not in every case.

Edited August 11, 2012 by qbick

[neonbulbs80]

@qbick : That would be great if SESI able to implement that. But for now the brute force method seems doing quite fine, since if I apply DOF or motion blur, the render time doesn't seem to be increased 3x or 4x longer compare than using MR. What I like about it is that it only add up additional 20% of render time when you have displacements as well. For production time this is one of the advantages I guess.

Anyway for the additional progress I finally finished my toy classic car modelled in blender and just finished surfacing and test renders. Next week I need to wrap up background interior and additional props for visualization.

Any crits and suggestions are welcome.

Cheers..

[neonbulbs80]

Finished up modelling the interior for the tin toy car. Next stop surfacing the room, final lighting, render and break them down to passes for compositing.