Luxrender doesn’t like .tif files.

All the ponies are skinned with .tif files.

So Luxrender gets the ponies just fine from Blender, but has no ability to display the textures.

Luckily this is not the end of the world, I can still use the ponies in something else… theoretically. I’ll get back to you guys on that one.

Cheerilee’s Gmod Pony models for Blender baked for 5 hours, 56 minutes, and 15 seconds in Luxrender. The image achieved  2.17 Thousand Samples per Pixel at 1080p (1920x1080).

This was done with Blender and Luxrender like the last two, and with the same hardware configuration, with the exception that I fed Luxrender all 4 of my CPU cores for 4 hours of that render time. This was Bidirectional Path Tracing with Metropolis Light Transport again, with the one change being the tone Mapping kernel was swapped to MaxWhite from Auto Linear in the last image, this made the image less grey.

That said, I learned a couple things from this run through:

First I learned Luxrender 0.08 (the current stable release) does not properly support OpenCL GPU computation, so my graphics card wasn’t receiving its proper share of the work in that last render. As a result of that, I set this run through to be purely CPU based, and fed the extra core to make it faster overall.

Second I learned that rendering in 1080p is something I do NOT want to do on a regular basis. It took FIVE HOURS of baking to create this image.

Third, I have no idea how to properly export textures into Luxrender along with the models from Blender.

Fourth, I learned that Luxrender does not converge an image well with high-poly models. If you don’t understand what that means, look closely for the white dots in this image. They only appear around the models themselves and appear to be artifacts of low sampling rates and non existent caustics that the renderer hasn’t sampled out yet. For this image though, which was merely a proof of concept for the quality of the models themselves, five hours was more than enough. Besides, theoretically, what is left could be simply blurred away by a soft blur. Based on what I’ve seen and heard, 44 thousand samples per pixel would likely eliminate those dots, but it would require me leaving my computer on for 5 days (121.658986 hours) if I can’t get my graphics card working with this.

Overall this was a learning experience, and definitely taught me quite a lot about these tools and this particular rendering algorithm and how they both behave.

I apologize for the camera positioning, I did not start this with the intention of baking it, but ended up doing it anyway for the sake of time on my part. I realized I would need to check the quality of the models at some point, and this camera angle was good enough for Celestia, who was the main focus of this test.

Incidentally, tell me if this works as a decent desktop background.

EDIT: 1080p version here!

The same cube and light source from the last scene, this time with a simple plane underneath it. The plane creates another surface to scatter, refract, and reflect light from the source. This makes the cube something other than pure white, and also provides a method for lighting the back of the cube.

Rendered at: 3.55 Thousand Samples per Pixel

Render Time: 1:52:10 (hours:minutes:seconds)

Render Method: Blender(Luxrender) OpenCL Metropolis Light Transport + Bidirectional Path Tracing. On Nvidia 9600GT and Phenom II x4 965 (3 render cores @ 4.01 GHz a core.)

Time for something more advanced, perhaps?

If you ever wondered what a simple cube would look like if there was nothing else in the entire universe except a lone light source, it’s this.

This is a light source, a cube, and nothing else. There’s no subtlety or definition to the light at all because it’s completely unobstructed, it’s not being reflected, it’s not being refracted, and the surface of the cube has no definable characteristics. So there’s only the light sides, and the dark sides, nothing else.

Profound isn’t it?

Luxrender isn’t working well with Blender. That is to say, rendering the scene produces absolute nothing but a black screen, which clearly means it isn’t working. I believe the world is lit, but even if it isn’t a path tracing program should not finish an image in 1 second… should it?

Uhg, installing Blender and Luxrender has just been unnaturally complicated, the instructions were clearly written by engineers who assume everyone else knows what they know. In any case it looks like I’m missing some sort of ply file that I can’t find anywhere, but someone assumed it would be really obvious to find it. Searching for it on my computer returns no results, and trying to find it otherwise seems to be an exercise in futility. So, at this point I’m willing to give up, uninstall some of this, and reinstall a version with Luxrender already built in, I just hope the configuration of it is not anywhere near as overly complicated.

Meanwhile, I’m using Blender because the Gmod pony models are actually textured and rigged properly for it, and Luxrender is a pretty high quality path tracer that can take advantage of a GPU and CPU at the same time. That should mean that, theoretically, a render style Celestia blog can dish out answers to questions in a few minutes, with most of that being me typing the text response and posing Celestia in a scene. The images themselves don’t have to be large, so they can be easily created in a few seconds time.

As it stands this is still far away, as I don’t have additional 3D models of castles/buildings/interiors/random objects to render Celestia with, so if I start doing this it will be in generic skyscapes for quite a long time. Add to this the fact that the Celestia model has her hair flowing in one specific way, and a lot of these answers would look the same for a while, or just extremely similar with the main differences being the way the model is posed in the space and the lighting in the scene.

Still, it’s worth a try to do this, because lord knows I’m not getting any actual questions right now. Last one I got was Christmas Eve, and it asked whether Celestia likes robots. I hate to just abandon the blog while I figure all this out, but Celestia can’t do all that much without people asking her questions.

Octane Render is a rendering engine that uses path tracing to provide high quality realistic results on an entirely GPU based solution. All you need to use it is a CUDA enabled Nvidia graphics card. You don’t need anything else actually, the CUDA card is the only real minimum requirement. There is a minimum CPU speed measured in a few hundred MHz, but the truth is that Windows 7, Vista, and XP need these bare minimum just to install.

The renderer itself however, that costs a fair chunk of change. Right now the price is 99 Euros which is 136.4616 US dollars. That’s not including an Nvidia Geforce 9000 series or higher graphics card to actually use this with. So ultimately most people will likely not be willing to front the cash needed just to play around with a really cool rendering environment.

So for everyone who isn’t rich, me included, you can look around this lovely gallery of rendered images they have assembled. A word of warning, it’s MASSIVE. However, I find the images they have placed there are extremely impressive. All of the images in the gallery were created in Octane Render using path tracing. Imagine these images moving around, and then you’ve got a good view of what the future is going to be for video games.

So, why post this?

Well, other than the fact it’s awesome… I realize that all the noise coated youtube videos and highly technical descriptions are… well… very dry. Often times they can be very hard to understand at all, but it’s hard to garner enthusiasm for something that is presented in a dry, stale, and almost purely technical form. So here’s a good representation of what a ray tracing algorithm actually produces that isn’t dry, soulless, stale, and horrifically overly technical.

Not that I view any of the things I post here are dry, soulless, stale, or horrifically overly technical… but I know other people feel quite a bit differently about this sort of thing. In any case, I’m rambling, and there’s a ton of pictures you could be looking at in that link instead of reading this. Well? What are you waiting for? Go look at those pictures!

Ray Tracey just posted this up on his blog. It’s a video of a high detail, Path Tracing, Global Illumination, 3D renderer. It’s running on Nvidia’s GTX 580 Fermi GPU, it starts on just one, and then goes up to 8 at about 5 minutes in. The result is extremely close to real time at very high sampling rates with incredibly complex 3D scenes.

What does this mean? Well, considering the GTX 600 series {or whatever Nvidia ends up calling their new Kepler GPU} is supposed to be twice as fast as the 500 series at ray tracing and similar expensive computational programs. It means we can expect 4-way SLI to get these results on the latest hardware next year. Beyond that? 2-way SLI in the following year. Then finally on a single discrete graphics card in 2014.

At that point, 2-way SLI, a very popular choice among enthusiasts, will likely hit the magical real-time 30 FPS target. That’s 3 years from now, and only if more significant improvements don’t occur in the time frame.

That said, the video itself provides some very beautiful indirectly lit scenes. Turn it up to 720p and marvel at it for a while, it’s truly amazing to think we’re so close to this technology being put into place in interactive ways.

All of these things are ways of providing Global Illumination in a video game. All of them are currently considered impossible to implement. All of them are amazing. The idea behind Global Illumination is simple: simulate light. The answer to that idea though, that’s incredibly complex. Personally I’m a huge fan of Ray Tracing and Path Tracing as the results are simply lifelike shadows, reflections, refractions, and caustics.

These results take very little work on the part of the programmer in Global Illumination, but become incredibly complex for the computer to actually perform. As such, some of those images actually took minutes, hours, or even days to produce on a computer, although the coding took no where near as long as normal video game light models take to produce. The result of this has been a quest, since the dawn of computers, to make Global Illumination a reality. We’ve come a long way, in fact, you can practically say we’re here if you look at some of the demos by Ray Tracey. The truth is though, there’s still a gap that needs to be bridged before ultimate graphical realism becomes a reality. It wasn’t a large gap by any means, just 10 or 20 extra FPS in some systems would give us real time, interactive results in games with the visual complexity of Metro 2033 or Crysis.

However the gap is now gone, and it’s all thanks to an entirely unknown left field technology deceptively called “Image Based Lighting”. I say “deceptively called” because image based has the subtle implication that it’s light mapping of some kind. Which would be some sort of pre-baked texture with a cool lighting effect in it, right? Wrong. Turns out it’s Global Illumination in an easy to compute package. Not only that, but it slipped passed everyone in Criterion Games’ Need for Speed Hot Pursuit, and now in Turn 10’s Forza Motorsport 4 without even raising any interest. It got no praise, no massive applause, no Nobel awards for scientific achievement, nothing. IT CAME OUT OF NOWHERE.

So how does it work? How does this miracle algorithm provide Global Illumination to incredibly complex scenes at high frame rates? I have no idea.

The Wikipedia article offers little help here either. Its’ only external source is a download for a demo, not even a description, a full on download link. Someone needs to get all over this, now, because this was massive news that slipped past us all. Unless I’m missing something, and we’re being lied to, but if it’s real the implications are unbelievable.

More so, I’m frankly disturbed that despite my near obsession with global Illumination, and my love for everything Criterion Games does, I bought and played Need For Speed Hot Pursuit without even realizing what I was seeing. That says a lot of things, I’m sure, but still. I, for one, am stumped on this.