Project Widow Reboot "SILK"

[projectwidow]

Hello,

Some of you may or may not remember a short film project called “Project WIdow”. This was a short film that begun in 2009 as a technological showcase between Blender and Renderman, in our case we were using Blender 2.49 and Aqsis. At the time such technology was experimental but it functioned well, this was before Matt Ebb’s addon was picked up by Pixar, so we wanted to show that such a feat was possible for small and large scale environments.

This was a world wide collaboration between artists and software developers and had been worked on for a couple of years before it shut down in 2012 due to lack of updates and software issues.

Last year I secretly begun working on it again, this time using Cycles as the render engine. The reasoning was I was bored in between moving from one house to another and to pass the time messed with the assets we created using Blender 2.79. I was amazed at not only how well it looked but how easy it would be to convert everything to Cycles. So one thing led to another and here we are.

I have been decided to start posting on BlenderArtists to show some of the progress, as well as it’s history here and there.

Some renders I have done over the past year, quality is not great as these are more tests renders than final frames.

This is a composite test between a still background and the animated spider

Test render of the environment

2.79 Compositing test.
I have since upgraded my renderfarm to Blender 2.80, going to 2.81 soon to take advantage of new features recently added.

Composite test video

Another composite test done in Blender 2.79, this one was to bring out the spider webbing more.

I do a LOT of writing about this short on the blog that’s still here 10 years later but will try to keep the project up to date here on BlenderArtists as well.

Project Widow

This is the storyboard video

[projectwidow]

It has been some time since I worked on this but it has not been forgotten! It just got put on the back burner for a bit but once again I have been doing more and even have some final frames rendered.

So to begin with I decided that EEVEE would have to do as the render engine of choice, primarily due to the HOURS long render times and I simply do not have the resources, time nor money to shell out my own private renderfarm nor do I want to rely on a commercial renderfarm either ($$$). I have spent YEARS working on this and simply want to complete it before I am old.

I made this change after seeing some advances in EEVEE since the 2.80 versions, now I am using 2.93.

Off to some renders and screenshots!

This screenshot is from a shot about midway through the whole short, I am working on the easier shots first, while getting some of the harder animation done at a later time.
Here the train simply passes in front of the camera, and I applied a little noise to the camera to simulate camera shake.

Here you can see the end result.

This shot is in the beginning of the short to establish the general mood and atmosphere, environment that the whole thing will play out in.
I made use of the compositor and applied some lens flare, distortion and even film grain.

Here again is an opening shot, with more lens effects. I did try to get some Depth of Field going and used the compositor to bring out the lighting a bit more. Again all of these renders are done in EEVEE.

The train tracks have finally been shaded to an acceptable degree, took me forever to get right considering that I had completed them in RSL (Renderman Shading Language) and had to replicate them in Blender. For years they stayed the same but a couple weeks ago finally got them to look right.

Here is the shader node layout.

Much of the shaders are procedural, there are actually very few objects that are texture mapped and the ones that are have a mix of both. I think the spider, our main character of the short, is the only thing fully texture mapped. Back in 2009 when this was a Blender/Renderman project we wanted to showcase the power of RSL at the time and using procedural shaders was the way to go. Time marches on and as explained in my first post (2 years ago…) this turned into a pure Blender project. Too many people laid their hands on this to let it go to waste so I am determined to finish it.

Anyways it is late, just wanted to post an update long overdue…

[projectwidow]

So today I have placed the first “trailer” for “Project Widow” on Vimeo, I tried my best to piece together the footage I do have rendered and composited so far, not all of it of course, just some of it. Hopefully this is of interest for some people.

[projectwidow]

I am here to give a small update on “Silk”, it has not been forgotten and I have not given up on it, actually quite the opposite, I have been very busy on it and have got quite the list of things done.

I wanted to share some renders (before and after composting), some screenshots and a test animation of a particularly difficult shot that I had to finish.

One thing I have done is installed Kitsu on my old system that is now more of a server, for those that don’t know Kitsu is a production tracker for things like this…

This has helped me out IMMENSELY! Not only do I NOT have to remember what shots I have done, in what stage they are at, I have a quick web page that I just have to go and look at. Before I had installed Kitsu I had to go through the files, see what stage they were in to remember “Oh yeah! That is what I was doing!”… very ineffective, time consuming and not very productive. Now all I have to do is look at the page, and do some clicks and I get an update on my progress, or I set an update on my progress. Now I could have done one better and installed the Blender Studio Kitsu addon for the production, but I would have had to set up an SVN server, get THAT configured, get the assets configured in such a way that the addon would work… and so on and so forth… no I decided to that I was already knee deep in the mire that I wanted to just have my directory structure as it is and just work out the shots, not worry about assets and even more server related technicalities. It was a days work just to get Kitsu running.

I had to fix the opening shot because of the small bit of geometry that you can see in the top right corner… this bit is actually way below the shot area, in terms of physical space and is lit because of the sun light I have in the scene. I am not sure what I changed in the camera area but as I was reviewing the scene this popped up in view… either way I worked on the camera animation a bit and now it is gone.

The shots for Seq 004 had to be rendered again because of a small technical glitch that I simply forgot about. The cables that you see drooping off the sides of the walls? The history of these is that they were built for Renderman, each object can have a Subdivision surface applied to them without having to do it in Blender, at least back in the RIBMosaic / Blender days of 2009… well I never did this to them when I rebooted the project a few years ago and they continued to be this way until I had the bright idea to apply the SUbD Modifier to them and lo and behold they look %100 better. Because I have switched to EEVEE for the final render I did not have to wait a few days to see the results, at best it was an hour combined, I hit render and did other things for a bit while my computer chugged out the frames.

Here you can see the spider in action, sort of, as it is traversing across it’s webbing. I have a better image somewhere but I could not for the life of me find it… I think… (I have a LOT of images) anyways…

Here is another shot of the same sequence just at another angle. I had to take liberties with the lighting to get the spider silk strands to look better than they did with just the default lighting setup, in this shot I have a light off to the right side of the camera to highlight the spider and another light off to the left to do the same but to also highlight the strands… otherwise they look black and very not “silky”… I just turned down the Diffuse and Volume lighting to 0.

This is an early render (it has since been rendered again for technical reasons) of Seq 009, Shot 001… this is a 2 second shot of the spider in the line of the train, about ready to be hit… The reason for the rendering again and of course I don’t have a PNG of that render is because the silk could barely be seen and the camera in ALL other shots except this one had 6 blades, so the Depth of Field looked very different than the rest.

This was actually taken tonight during a testing, this webbing is actually a cloth simulation. The train comes and rips it apart, I didn’t have to do anything other than make a series of boxes (that are invisible to render) as collision objects, attach them to the train objects and then animate it moving at high speed and the simulation itself tore the webbing apart. It was pretty cool.

This shot of course took a lot of work to get done. The first shot I did of this didn’t have the oomph I needed to show that maybe these move the air and are heavy. Of course this is all just for visuals and in all reality I have seen plenty of trains not blow dust around like this (snow on the other hand is different I think)… I had to create a cone emitter object, parent it to the train object, then through trail and error made a simulation of dust by creating large icospheres and then using an animated volume shader to create the “dust”.

Here I have a quick video of that shot.

The shader for the above shot.

Another thing I wanted to bring up that might anger some Blender users is that for the longest time I had been using Blender’s Compositor for compositing my shots. Well I decided that I wanted to try to Natron again for this short. One reason is that I wanted to get my hands dirty on a true compositor and I felt that while Blender’s is powerful enough for what I am doing, it lacked in certain areas. I am sure some of you will be offended and I am sorry if that does… and yes I know that had the compositor not been added to Blender then certain things like MultiLayer OpenEXR files wouldn’t exist. Trust me I know, I was there many many years ago when Blender first GOT the compositor, it was a glorious thing to play with. Plus I am not the only one that uses other compositing software for the final product, I see many of you guys using other software. Anyways off my soap box lol…

I unfortunately do not have any screenshots of my composites yet, I will in the future I think, considering how LONG this film is taking, but you have to remember I am just a one man band working on this now.

[projectwidow]

Well things have changed, I am not satisfied with the results I am getting with EEVEE, mainly with volume lighting being visible behind mesh objects and no matter what I try to do I just cannot seem to get rid of it. Either way I have reverted back to Cycles and realized that I had been doing renders all wrong, which is why I was getting fireflies and noise.

So to explain what I was going through. Here I have a light that is behind cylinder mesh, it should not be visible but it is.

Here too there is a light that is behind the mesh and it is visible through the mesh. Not sure why this is happening.

So that made me switch back to Cycles and I had done some testing to see if I still would get some noisy renders. I also read up on it a lot, trying to make sense of what I was doing wrong. Some of the images in my previous posts are examples of noisy images.

Below is a test render I made with motion blur, I had clamped the indirect lighting a bit and filter glossiness as well. I also had some volumetrics in there as well but I think in the end I will be removing fog just to shave off some render times. There is also denoising done to the render.

I also did some more tests that I saved as EXR and for the life of me cannot seem to open in Blender to resave them as PNG but I am sure it is just me. Anyways off to the next subject - the renderfarm.

The Widow farm (as I call it) is the next version based off of the experience I had a couple of years ago when I rebooted this short film and was trying to save some time rendering. It took a couple of weeks to get down and unfortunately I did not document my process but I still remember much of it. The renderfarm is based on CGRU / Afanasy, which has been around for some years now and works well. It installs nice and configures fairly easy so getting THAT part done is pretty painless. The bulk of the work though is getting the farm to use a network file share and making sure the right permissions are in place. In the first version of this renderfarm I used Samba because I was thinking of using some Windows laptops I had laying around to make use of them (bad idea as they aren’t designed for such use) and getting this to work was a pain. I also had just 2 Ubuntu systems. Now I have 3 hah! If I can afford it I will try to get more but at the moment I have 3 systems that are available to use. My workstation and 2 older systems I have, one being the Kitsu server. Not ideal situation to use a server as a render node but I have to make do with what I have. My other system is an audio workstation I have running a variant of Ubuntu 16.04 and haven’t really made use of it recently in over a year so I THINK it can be wiped and redone as a render node at this point.

I will be using NFS as the network file share, it is said that it is easier to use than Samba. I have yet to actually install and configure it. The previous version of the farm took a couple of weeks to work out the kinks and issues due to the Samba configuration and file permissions drama. Hopefully I won’t have the same issues here but am prepared for it.

In all I think I am making the right choice in going back to Cycles for the rendering, I like EEVEE and it’s fast but I just did not fully like the renders I was getting from it and I was making sacrifices for the sake of speed. Now that I know a lot more of Cycles due to actually reading and testing my shots, I think the end result will look a LOT better.

Oh and also in my testing I shaved off a LOT of render time, I also started to use the OpenImageDenoise which makes my renders look a lot better than the NLM I was testing a while back, which looked awful in my opinion.

I think I will stop there, I want to try to get some more done before I go to bed it is late.

[projectwidow]

Showing some newer renders recently, just testing each shot so I can see what to improve for either better performance, lighting, or any other reason for that matter. Below is a newer render of a shot, on EEVEE looked not so great as seen in a previous post. However with Cycles looks much better.

One such render I noticed that there was some “broken” geometry, actually it is where the tunnel section repeats and the geometry for the wires didn’t connect.

You can see it in the left side of the frame, very clearly didn’t match up.
So I took the original model, deleted the offending geometry and simply made new wires.

Another test

Others I am not so successful on… here I have a lot of green fireflies that I am trying to figure out how to reduce or get rid of.

I think I need to tweak the lighting? I am not sure but it’s going to be addressed either way.

In the mean time I have been checking out EBay for cheap servers to act as rendernodes, I think by winter I could have a relatively cheap farm that would at least help render frames out. My current rendertimes of tests I have done of this on the small scale farm has taken at least a day and a half for a few hundred frames each scene, so not bad per se but still would like the boost, or at least free up my workstation. I have also found 16-24 port switches for like $30, so networking would be cheap. Maybe not the electric bill but that’s another problem for another day.

[projectwidow]

Another update.

Did some work to the main character, the spider. Mainly in surfacing, which I added a SSS map to shader.

Nothing too complicated in the spider shader because much of it is texture maps. Back in the Blender 2.49 days it was much the same as well. Of course back in 2009 we were merging Renderman and Blender together so it was up to the RSL surface shader to make use of the texture maps.

Since this is using Cycles on Blender 2.93 now I decided to go the Principled Shader rather than try to roll my own. I tried but the results were not as good, or in other words it looked ok but to me it looked off. Maybe I have just spent hours looking at the same thing anything different just won’t match up. I don’t know.

Anyways, I am still chugging away at this trying to get shots either finished or started. I am planning this to take longer than the end of the year, yet again. I am not loosing hope, in fact the opposite. Now that I have a pipeline of sorts built for animation, rendering and compositing it is just artistic driven. Now I gotta work on the actual hard part. Recently I was working on a physics simulation that just wasn’t working right, so the solution is to just animate by hand. Luckily it is just one object. I have rendered some shots again, this time in Cycles on my still 3 system renderfarm. Life though gets in the way so I have had less time to work on it as I would like to.

At least I did the last bit of work on the Spider shader now rather than halfway through the rendering and changing it.