One Man, One Month VFX Sequence- Making Of Horrors Of The Skies

Intro and Project Synopsis:

On the 13th October I started working on a 30 Second Sequence about a RAF World War One plane been shot down and crashing into the ocean, I had until November the 10th to finish the sequence. Not only was I alone on the project, not only was I doing everything (Not Just FX) I even had to model the plane and texture it! I knew this was going to be a huge undertaken, and I knew some of the quality would suffer due to the time, but here is the Making Of the Sequence, Horrors Of The Skies.

I always really wanted to make a WW1/WW2 Dogfight Sequence, so I kinda saw this as an opportunity to test my skills, so I can think about doing my own sequence in the future. This is the first VFX Sequence I have ever done!


Note I plan to make a Full Breakdown reel when I get time.

PRE PRODUCTION

Making an Animatic in 1 Day

With one day to make an Animatic I setup four scenes, one for each shot, using low res assets and basic PFlow systems I blocked out the animation of the Plane (Sopwith Camel) and the PFlow systems blocked out the Explosions,Engine Fire, and Water Splashes. The Animatics came in very useful throughout production, and they got my ideas for the sequences into 3D from my head. Some Ideas I had did not work and if I had not of tried to Previs/Animatic them then I would of wasted time which I could not afford here. I rendered the Animatics through QuickSilver Renderer and then edited them together in AE, also I all of the Animation and Camera Animation from the Animatics for the base animation/setup in all four shots.

Animatics In Progress For Shot 4

4 Day RnD Of Plane Destruction and 3D Volumetric Clouds

I spent around 4 days of RnD before Production started, my main worry was about clouds, I spent around 2 days Cloud tests with Afterburn which I now own, In the end I didn't use Afterburn at all expect for Shot 1 due to the time constraints. But I used Afterburn to see how detailed I could get 3D Clouds, and the results were not bad at all, I think if I had more time on the project I could of used AB and it would of been really nice. In the End most of the clouds are ethier pre rendered FumeFX Clouds or Stock Photo Clouds on Cards, and the Sky-domes which have there own clouds.

Test Shot For Clouds ^^

Another thing that got scrapped due to time constraints but was part of the Pre Production Days was a PFlow tool I started working on called "DogFightTool" Which was a system I started working on to have Background planes dogfighting each other to add more depth/detail to the shots, It was based on lots of test events so that if one particle (Plane) flew close to another one it would get shot down, or if it was in the air for so long it would get shot down, then to get them chasing each other I used wind and find target events. Sadly I did not have time to develop this further but may do in future personal projects.

Very Early "DogFightTool" Development

The last day I spent doing Plane Destruction RnD, I initially was thinking for Shot 1 I would use nCloth in Maya to get the tearing wing, however again due to time constraints I could not, due to the time it takes to tweak the sims and then reexporting from Maya to Max through Alembic. In the end I used RayFire and rather than just getting a normal RBD sim I added bending/Flexing to the fragments by Keyframes using FFDs and Bend Modifiers to add a more realistic breaking/snapping feel. I stayed far away from Voronoi Fractures. See More Below.

Wing Destruction RnD- Maya nCloth

Wing Destruction RnD- Rayfire (Max) Plus Keyframe Bend/Flexing

I looked at reference's such as King Kong (Weta Digital 2005), RedBaron (Pixomondo 2008) and SuckerPunch (Pixomondo 2011) as they all have Biplane Destruction in, It seems Weta used a more Rigid/Soft body approach for the Biplane Destruction in the New York Sequence, where as Pixomondo went for a more cloth based approach of the tearing wings. Ideally I wanted to go down the cloth route and then add internal details like the airframe when the wing gets ripped off, but I went with a more Deforming Rigid Body sim like in King Kong.

Amazing Example Of Weta Digital's Use Of Deforming Rigid Bodies on King Kong (2005)

Modelling And Texuring a Sopwith Camel in one 36 Hour Day

As you will know I am not a Modeller or a Texturer and dont have many skills/interest in that area of 3D although I do model a bit for FX when creating Proxy Meshes. The Acual Asset of the plane/Sopwith I worked on when I was in College so I sorted out the bad topology in some areas, I then just went in and added lots of detail to the plane, I used a Free Asset by Virt3D for referencing.
I then recreated the UVs in Maya and sent it to Mudbox to Paint. I painted Diffuse Maps for each part of the plane, also Dirt Maps, and some very crude Bump Maps.

Once these were done I brought the Plane Back into Max and created a rig for all the moving parts on the plane including the Pilot. (The Pilot I did not Model).

Texturing The Sopwith In Mudbox

Rigging The Sopwith In Max, With Basic Controls

Pretty please with the asset for the amount of time I had, its nowhere near as good as it should be but I really dont have the time or skills in this area of 3D.

PRODUCTION

Making Of Shot 1 and a Huge Setback on Day 1!

I actually have an article Below about the making of Shot 1, but I will explain a synopsis here,
So Day 1 of production I faced HardDrive Corruption on my C Drive, which took down my DNS causing DNS Failure, after 10 Hours of frustration I had to do a Clean Reinstall of my OS which was soo annoying! More so resintalling all my software and Licenses. However it was nice to get some of the junk off my HDDs!

Shot 1 has to establish the look of the Sequence so it was important to get it looking as good as possible, I used the Animatic Animation for a base for my Hires Animation, I then added some little wobbles of the plane and animated the flaps/wings to move responding to the motion of the plane.
Also the German plane was animated but did not have a rig like the hero plane as there was no need to waste time.

Then I manually fractured the plane wing by cutting the vertexs up, and use Rayfire with activate by object to animate a box to pass over the wing causing it to break apart in the frame ranger I planned to have the explosion in when the plane reacts.

Plane Destruction Sim

For the Explosion I knew that MG Rounds to the wing would not cause an fireball, however the shot needed it to sell the dramatic effect, I went for an Airburst Style Explosion with Smoke Trails, Using Fume FX Ofocourse! With Particles for the emission, the Actual Plane and Fragments do not interact with the sims becuase I would need bigger grids and I did not have time.

Shot 1 is the only shot that has 3D Clouds in, and most of them are cropped out in the final shot anyway, Using Afterburn and a Basic PFlow system emitted on a plane which shapes the clouds, One thing I needed to make sure of was not to over do the turbulence in the clouds, Yes its adds detail which is needed BUT if you add to much, it looks really crap!

To Much Turb on AfterBurn Clouds.

The rest of the clouds are cards and then the Skydome in the background. For all of the shots, I could not do proper renderpasses, I was relying on Beauty Passes for nearly every render pass (Not FX).  Composting was done within 5 hours, the aim for the sequence was to be quite flat and desatruated.

SHOT 1 Breakdown and Comp Work In Progress

SHOT 01 BREAKDOWN REEL

Making Of Shot 2 & 3

Shot 2 and 3 were done together based on the same Scene Files, apart from different cameras. The main problems I faced on these shots were due to the camera been so close to the plane and the engine fire.

Again the animation was based of the animatic, and then more details added with the hires rig.
I spent one day before Production for the shots setting up Flak scenes. Based of the techniques of flak I did for WarThunder I setup 5 different grids. All were quite lowres as unlike WarThunder, these flaks would be flying past camera rather than exploding in the sky as the camera moves around them. This also meant there would be lots of motion blur on them. They were using a main PFlow emitter for a small blast, then mushrooming smoke trails, and then thinner smoke trails. Then using Fume Colour Caching to get the smokes mixing together.

Two Of The Flak Sims^^

The Engine fire was done using a grid attached to the plane and using Wind to fake the velocity because the grid would have to be huge if not. Again I did this on WarThunder for the B-17 Engine Smoke. Using Particle Emission from the engine, and low velocity so the wing would "Push" it back. I got some feedback from a friend after the shot was final saying the smoke looks out of scale compared to the flak flying past the camera, I totally agree with this, The flak serves as a speed reference to how fast the plane is falling, The Problem is I did not want a huge scene, so By using the flak I wanted to fake the scale.

Initally the plan was to merge the Flak grids into the scene as "Preset Grids" however in the end I just rendered each flak elements from two different angles and then hand tracked them in comp flying past camera. One really good idea I had was to put lots of these flak renders in one comp and have them explode at random frames, and then use this comp in the background of Shot 2 to add more Depth to the shot. I think it worked really well.

 Shot 2 Scene Setup

RePos'd Camera For Shot 3, Using the Same Scene Files As Shot 2

I also then use a PFlow system to emit sparks coming from the fire, To add more detail to the engine fire sim, I did some post sharpening in comp also.

Rendering With Pretty Simple Setups In V-Ray

Lighting and Rendering was done with V-Ray as were all the shots, Using A HDRI plus some V-Ray Lights in some areas. Again the comp was very flat and desaturated.

Shot 3- Close to Final

Making Of Shot 4

Shot 4 was going to be no doubt the biggest shot, It was very good fun crashing the Biplane into the ocean! I had 1 week to do the whole shot from Scene Setup all the way to Final Comp, I had pretty much 12-15 Hour days for a week for the whole 5 seconds of the shot, and in the end the shot was only 4 seconds due to the edit!

For Refrence for this shot I looked at a really old but awesome Game Cinematic that was done in 2003 for Medal Of Honour Rising Sun, Called "Welcome To Paradise Soldier", The last shot is of a Japanese WW2 Zero Crashing in the sea, and for 2003 it looks awesome, Back then there will of been no Fluid sims ofcourse, Also for Reference I used the Deep Gold C130 Crash Shot.

Shot 4 Animatic

For the Animation I had to get it dead on, not only for the scale to look right, but also it needed to play well to camera, Theres not really much to say about the animation other than I had huge headaches getting the plane to look like its sumersaulting without making it look out of scale. I knew that the Destruction/RBD sims would be inplay as well as the Fluid sims so the Actual Animation was only going to be seen for the first 50 frames really. The RBD Sim was done by Manually cutting the plane up rather than a Voronoi Fracture as I did not want it to look like stone, Each piece of the plane was sim'd on its own to break at different times. If I had more time I would of wanted to go down a more compelx sim and add more bending and internal details etc. I did add some bent metal inside some of the frags for some small detailing.

Basic Small Cylinders with Noise to Fake Metal Inside The Plane

Once the RBD Sims were done I Keyframed some of the pieces to play to camera, like the tail to fly towards camera and splash. I then used an SD Export through the RealFlow Plugin to Export the final sim cache to a new max scene, where I could scale it down using a Helper/Scene Scaler so it would be the right size in RealFlow. Once the size was set, I spent a few hours modelling some crude Proxy Mesh for the Fluid Collisions on the Main/Hero Debris. Using the actual geomrties from the sim in RealFlow as the Collisons would be disastrous as with any Fluid Simulator the Collisons need to be clean. All the Proxy mesh is just lowres exturded boxes to match the shape. It doesn't have to be super accurate as the pieces are splash quite fast so you dont really see the collisions. The Proxy Mesh just gets parented to each piece of the Hires sim, then the Proxy Meshes get baked/cached to SD again and brought into RealFlow, The Final RealFlow Sim will ethier be scaled up once back in Max to match the Hires RBD Scale OR The Hires RBD Sim will be used at the Fluid Sims Smaller Scale. Aslong as eveythings scaled properly, some scenes like the engine fire, can be at the higher scale whereas the Fluid Render Scenes will be smaller scale, but it will all match up in Comp.

 See How Simple The Proxy Mesh Can Be^

In RealFlow I did lots of lowres tests first to check the scale of the fluid, as this is a main stumbling point in Fluid Sims. I knew that the main look of the sim would be coming from the Splash/Spray Sims and the Mist and Foam, but to get them looking right you need the core fluid to be working well. I used some Vorticity Boost to get the splashes looking more broken up and violent from the imapct of the plane. For the Collions between the Proxy mesh and the Fluid I used Shell Mode with a small Cell Size, it worked really well and was really quick.

Core Fluid Sim was around 25 Million Particles 

Splash For Mist Sim

For the Secondary sims, I used a main Splash Emitter using Dumb/Ballistic Particles that really shaped the dense splash cloud, I used a Noise Field to inject some nice splash turbulence to the sim. Then Instead of using VDB Mist via Voxels, I used a sepreate Splash Emitter, with Lower Gravity, a K Age (To Kill the Particles Off Quickly) also some direction wind to blow the mist backwards, this emitter was then rendered as Krakatoa Voxels and is a lot easier than the current VDB Pipeline through RFRK and Mental Ray. The Bubbles/Foam and Splash sims were also rendred through Krakatoa, I did plan to use Alembic Proxies, however the RealFlow Alembic file Stitching Script had some problems and I did not have time to patch an update. Rendering through Proxy has its benefits but the Krakatoa Render was pretty fast, and good enough for this project.

The actual Fluid Mesh, I planned to render through Frost and do WhiteWater Velocity shading, but in the end, It was taking to long to render, also I was cutting a lot of the mesh off anyway, as in reality when water is moving at the speed the Plane crashed into in, it would all be spray and mist rather than in a mesh like state, so I just bit the bullet and used a flat ocean plane for the ocean rather than blending the Fluid Mesh into an Extended/Infinte Ocean Plane. It worked well.

The Fluid Mesh Was Scrapped due to Long Render Times and Tight Deadline

But It would have been nice to have WhiteWater Velocity Shading.

The Final Sim Passes with the Krakatoa Voxels Mist (Bottom Middle)

Even the Core Fluid Particles Were Rendered and Used in Comp to add more splash density.

In the Final Comp I added a small island in the distance by 3D Tracking the Skydome Render with After Effects built in Camera tracker and then just used a basic matte painting with 3D Layers to put the island in. The Engine Fire, is actually the Shot 2 and 3 Grid merged into the Shot 4 scene file and just parented to the plane, the motion looks bad, but we only see it for 30 frames anyway becuase then the water would extinguish it.

SHOT 4 BREAKDOWN REEL

And there we go! Thats 4 weeks work and the Project is done, and was viewed at My Local College via Projection.

Project Evaluation/Final Summary

So due to the time limits there was a lot of stuff I could not do mainly to do with adding more details, and putting more time into FX,Lighting and Rendering. But for 4 weeks of intense hard work I am really happy with the sequence. I knew from the start this was not gonna be World OF Warplanes RealtimeUK Quality, But I really wanted to see how far I could go all alone within the timeframe.

Big Thanks To- Middlesbrough College, Franck Dhotel, Trevor Brewis, Zoe Lewis. Also Thanks to Graham Collier and Cris Scubli  for there feedback throughout the project.

I plan to do a Breakdown Reel When I get time, If you want to watch the Final Sequence, its not on Vimeo but here:

I also plan to release my Personal Production notes once I edit all the bad spelling out of them, so anyone whos interested can learn from my mistakes if they take on a gig like this!

Thanks A Lot Guys!- News Projects Coming Soon!