Making of details

One of my last posts if not last. This is lion sculpture I did for this project few months back. I did not based it on any reference image or so at start. It was more of a free flow work, just to see where I get with it. Once I got some shapes and edges I decided to make it lion alike - but not exact... I spend quite a lot of time on making this model - would be probably less if I had some concept sketches but well...

At the end of the sculpture I remodeled the base to low poly, baked all textures and exported to max. The end result can be seen on one of max animations - the head is attached to building pillar.

 Subdivision 1

 Subdivision 4

 Subdivision 7

Post production one of buildings in after effects. My post work most of the time is made out of layers that I then use to bild up my final beauty pass. I prefer to do it this way because it give me a lot more control over my final result and let me achiev effects that I could not achieve using just a beauty render.

So lets get it started. As first layer I pick up GI- I use this layer because it let me see all that is in a scene and control overall tint of objects. It is worth to notice that I cant completly change colors of buildings because their bounce reflection would be different from new one. In that case it is crucial to be as accurate as possible in render application or do only small changes in post to collors.

 This is a pure GI, it looks quite ok not much to change except for the hue of the lion heads

 And here is GI with lion heads hue changed

 Darken down the floor and hue correction

 Here you can see GI + Light pass

 GI + Light + Reflection pass now its worth noticing that each light, reflection, refraction etc passes are being retouch and tune up to my neews

Some more retouching diffferent areas, window, cement, metal and so on. Tiny changes nothing too big.

 And here we have final image, below you can see breakdown of AO pass and light pass

Since I am a big fan of keeping all as physically accurate as I can I blended my AO with Light pass. I did it because in reality when we look at direct light there is no shadows where light hit, however the extra AO pass apears everywhere else where light pass dont hit. There fore I ADD my light pass on top of AO and decreased it to 25%. Thanks to that AO sems to disapear in direct light areas but it still remains in shadow giving me an extra details.

I have also used AO pass in some cases to add extra dirt and discolorations, using different noise maps to colorise AO pass and then use it as color mask for final image I was able to add extra level of dirt and discolorations.

Here you can see my light pass. I was unhappy with a lot on this image there fore I started on tweaking hue and brightness of my objects untill I was happy with it.

The image above is getting an extra relight from my post work. I have used normal pass in world coordinates to produce a RGB based image. I then used it to add an extra layer of light on my building to boost up the sunlight. I used this technique in many situations such as adding extra light to glass. It is not really physically accurate, but because the glass is 100% transparent and clean in CGI it does not catch any light, where as in reality it would be dirty and grey which would make it catch a layer of light.

Above you can see my final light pass, I believei s roughtly where I wanted it to be, the only mistake I did was to put the two mid ection blocks too bright making it a bit too unrealistic - will need to fix it.

Nvidia GDC 2012

Just stulmle upon some news in GPU world. Nvidia did presentation on 15 may at GDC 2012. http://www.youtube.com/watch?v=um6rMjwSNdU&feature=relmfu

Quite epic stuff if u ask me, new features such like smx, hyper Q and dynamic parallelism. Entire presentation is quite interesting full of news. The one that make my jump is CUDA 5 - which is enabling GPUS to acces "A" gpu acces memory form "B" gpu meaning 4x 4gb gpu will give up to 16 gb of ram to use. Which is pretty much avarege now days memory that a scene would use. Meaning GPU could be used to render images- not to mention 4x GPU will produce insance speeds for rendering...

Recently I started to experiment with GPU on iRay and Vray for more production based work. Since my workflow has changes by my proffesional work I now have a reason to use iRay which would speed up my render quite a lot in simple tasks that can be done on GPU.

I have already asked chaps at Chaos Group Vray makes to see if they know about the juicy news and they are going to look in to it, making me wanna dream that the answer will be "Yes we can share memory now"

Next months will be very interesting in GPU world...

GPU vs CPU.
Right so this is part of one of my main topics I set up for this course. I wanted to see if I can create a movie quality assets and use them in games and GPU based production engines.

In this simple test I have focused on IBL lighting between Vray and iRay in GPU and CPU mode.

The test I pick is very simple I used one of my buildings that I have moddeled for my project and applied a grey material so that I can diagnose the quality of lighting, shadows, noise and so on. This is very simple test but it will tell us a lot about how well does an engine precess lighting and sample image.

I have used for this purpose a

2x xeon E5530 @ 3.7ghz

GTX 480 - 1.5gb

GRS 8800 - 640mb

Each test render was set up to render only for 1 minute. In right upper there is a statistic for Vray saying how many patch has render achieved in 1 minute time. Unfortunately iRay dont have such features so we have to compare them visually.

I tested Vray only on CPU and GTX 480 because the GTS 8800 did not work. The result you can see below.

As you can see the GPU render is almost 4x quicker than CPU render, also the IBL lighting is very close. However there is a difference in shadow details in small areas, it looks like a ambient occlusion pass is being added or the shadows are simply more/less accurate...]

The second tests were made using iRay. I have tested them using, CPU only, GPU gtx480only, CPU + GTX 480, CPU + GTX480 + GTS 8800 in case of the last test I had to lower the resolution of HDRI from 4k to 2k in order to fit in memory.

As you can see the tests are quite simmilar to Vray, with the difference in shadows. It apears that iRay GPU match vray CPU and iRay CPU render in shadow details. Quite interesting that Vray GPU produce different result which now sems to be inaccurate. However the issue with iRay is that it produces a checker box alike noise which looks unnatural. The noise can be best observed on iRay CPU example.
Both of these tests worked quite well and did show the bennefit over CPU rendering, however this example is still to simple to be conclusive for using GPU rendering in real production. There are a lot of features missing such as, render elements, hair and fur, volume effects and so on. Non the less they do get them time by time. One of recent additions to Vray RT render is displacement. Unfortunately its still unsuported on GPU mode where as iRay can do it.

Another quite interesting point to note is how well is render utilizing GPU. I have run a lot more tests in past and I realised that in quick renders if I were to use GTX 480 and GTS 8800 the render took longer than using single GTX 480. Since the difference was about 18 seconds to 16 seconds, I believe that this could be an issue of loading the secene to another GPU, or how well the software manges the Cuda cores on GPU cards.

Dare To Be Digital...

A little making off, in this project I had to deal with 3ds max physics, video montage, post production and so on.

The idea was to create an entry to our video that would be a theatre curtain that opens and there you would see us. However before it would open we would see our logo and company name. The idea I had in mind was to have a game name pop up from top like a puppet on string, wiggle around for a few seconds and then got pulled back up. Also before that would happen a company name would display.

Now the physics was quite easy part. I modelled a simple quad based mesh – it deformed nicer than triangle based mesh. Once I got it done I rigged it to behave like cloth and attached it edges to around 5 helpers. Each helper was animated to move at its own unique time to simulate opening of a curtain. Once I had the animation of helpers where I wanted it to be I started to sim cloth. I run in to various problems such a scale issue, cloth parameters issue, quite big stretching of cloth and so on. It took me quite a while to polish the animation and get it where I wanted it to be. Once the simulation was done I have baked the key frames so that I don't need to simulate it again. However I only had simulated half of the curtain, the right half I just inverted and slightly offset the key frames.

The next part was to drop a Puppet Show sign and deal with animation. Now u would think this would be easy and all I had to do is to use a rope modifier in 3ds max, however that was no the case. 3Ds max 2012 and early version are a mid way through being evolved to GPU based physics from CPU. Because of that a lot of features were still missing and I did mist the GPU rope based modifiers. In that case I had to hand animate the rope. There were few cheats to get the GPU physics to simulate the rope effect however I had way to little control over the effect to use it in my final scene. After hours of tries and fails I ended up animating the rope by hand – well not really by hand I had a little helper script that simulated a wiggle effect on bones. What I did was to create a 6 bone straight “leg”, which then I used a source for my rope mesh. Once the bone rope was done I animated the first bone to move the way I would like the Puppet Show to appear. Once the bone was rotating and moving I set up the script to animate remaining of bones. What it did is to calculate how much its being moved and then add a bit of offset to make it look more natural. Its quite hard to explain, the best example is to grab a chain and just hold it by bottom and drop the top of it. It will wiggle and hang around until it get still and straight again. This is exactly what script did and all I had to do is to spend some time on re siming it until I had the sort of animation I was looking for.

That was the end of my animation process next part was to deal with rendering. Now I wanted it to be very quick and efficient, since I was on deadline I did not use any GI. I decided to have a 4 lights in my scene.


1 – main spot light for curtain
2 – A company name light that would project the company logo on to curtain
3 – A spot light for puppet show logo
4 – a fill light to simulate GI

Now you ask why so many lights. Well I needed to have full control over the lighting aspects in post production. To be able to tune the light blend of company name and puppet show were quite hard to do in 3ds max and the best way to deal with it was to control them in post. I could enable and disable any of the lights at any time. Thanks to that re lighting the animation in post work was a piece of cake. A specially in 32 bit where I could extract the data from the darkest pixel. Below you can see just a few of the render layers I was working on. Aditionally I have rendered a flare, smoke, dust and other effects

Once I loaded up all the light layers in After Effect I begin to composite them together. It was quite a fun part because the beauty shot out of 3ds max look very bad. There were a lot of over burns on puppet show logo because it was so active and moving so much that the lights did not really work on it. I had to use fill light as well as spot light and logo light to blend them together and create a nice lighted logo. I wanted to keep the cartoon alike look with a bit of light direction in this animation. At the end of day I used a lot of extra footage as well as pictures to simulate lens flare, smoke and so on.