January 14, 2008
Interview with Mark Stasiuk – Fusion CI Studios
National Treasure: Book of Secrets
Mark Stasiuk is known internationally as an outstandng RealFlow artist and expert. As co-founder and CG fluid fx supervisor at Fusion CI Studios, Mark has developed fast proprietary methodologies for dealing with complicated, high-volume fluid fx and has custom designed his studio’s streamlined, efficient fluid fx pipeline. Mark ‘gets it’ when it comes to the physics of fluid, he has a PhD in fluid mechanics and has worked for 20 years in the field. He also works closely with the engineers at Next Limit (makers of RealFlow) to constantly improve the software and resolve production issues. His scripting was instrumental in moving RealFlow from version 3 to 4, and he wrote the scripting guide in the RF4 user manual. Mark is currently advising Next Limit as they develop RF5. Mark presented RealFlow's merits to the technical committee of the Academy of Motion Picture Arts and Sciences on behalf of Next Limit. Mark's efforts contributed to Next Limit's recent Academy Award for Scientific and Technical Achievement.
So we last heard about Fusion’s work in The Guardian (Touchstone Pictures), when you worked with Flash Film Works developing methods for creating wave crests, white-caps and interaction between boat geometry and a stormy ocean surfaces. The results of that were phenomenal. Has Fusion had any significant developments since then?
Fusion has done a lot of development on our pipeline tools through commercial work. Commercial projects often have very tight deadlines, so you have to be really efficient. So we took a lot of our methodologies that were fast at doing unusual things with fluids and building them into streamlined tools that make it easier for artists to focus on the artistry and not the ‘geek factor.’ We’ve also done a lot of r&d, especially on fire where we’ve taken our initial crude models of fire behavior and made them faster, more stable and able to handle larger scale, more turbulent fire behaviors as well as to generate fire from moving sources and from fluid, like burning oil or a jet of fuel from a flame-thrower nozzle. And we’ve also, for the commercial work, developed a variety of methods of targeting fluids so we can take shapes, whether they are geometry or fluid based, and target a sim to those shapes. This gives us a huge amount of control to achieve some of the quirky behaviors that commercials demand (see our website for the Target commercial with Radium, and for Coke with Buck Design).
When Fusion collaborates with our client vfx studios, we bring to bear this entire library of these developments and methods which probably reduces the r&d cycle by a factor of ten. My work on National Treasure: Book of Secrets (Walt Disney Pictures) for example, the turbulent river methodology would be something that would take months to develop if starting from scratch. But we were able to adapt methodologies from Fusion’s past work to get a version ready for a comp in a few weeks time. It’s a huge time, money and angst saver for the studios we work with.
Speaking of National Treasure: Book of Secrets, you worked with Asylum VFX in Santa Monica on that, what fluid fx work did you do for them?
Jeff Werner, head of CG at Asylum and Jason Schugardt, vfx supervisor, got me involved to work on three major areas:
In the film, a city goes from being submerged to draining and then filling up again. The cg elements had to match perfectly with the practical elements. The cg elements probably took up about 70 percent of the shots. ie. a lake-like surface in the city, water streaming down the sides of buildings & pillars, gushes of water, spurts of water, and high pressure bursting through gates.
Asylum had a team of artists working on the fluid fx for National Treasure, so they needed efficiency tools for the fluid pipeline. The team’s lead artist, Gunther Schatz and assist Andy Cochrane, are experienced with effects and RealFlow, but methods for achieving some of the effects weren’t known yet, and there were just a lot of elements for them to do. They had nearly 40 shots involving fluids, some of them involving 10’s of elements each. So I developed tools using python scripting that allowed us to do several things a lot fast than usual. Things like automatically setting up scene files and automatically launching sets of sim files on a render farm with just the press of a computer key for the artists. It increased the speed of the work by an order of magnitude. So artists could then focus on making the water elements look good, rather than worrying about the technical aspects of debugging scenes or setting up new scenes. It was a fast development cycle, with me creating tools and Andy and Gunther immediately testing them and providing feedback for revisions.
I also developed the ‘raging river’ simulation. This involved setting up a fluid flow containing dynamic moving vortices which we could control the distribution of. At the same time, we could run a high resolution sim that was multi-threaded. So rather than looking like flowing chocolate or lava, the flows would look like raging rivers. The results were fantastic. This is probably the first time this has been done with RealFlow to my knowledge -- the first time a fluid flow from RealFlow was made to look like a super-turbulent raging river flow.
On top of that, in the river sim, we improved a previously developed method for adding foam, making the foam creation much faster, more controllable and more realistic. I also set it up so that the river simulation was fast -- for a few hundred frames the sim time would be overnight, which is really fast. Adding the foam layer would be another few hours on top of that. So in a 24- hour period it was possible to come up with a completely new version of the core simulation. After that we could add in layers of spray in the splashy areas to supplement the violent look of the flow. Gunther really ran with the resulting elements, finding very clever ways to light and render them as well as re-use them in other shots.
We also developed an image based force-field system because there were a lot of shots with water running down highly textured rock and brick surfaces. But the geometry didn’t exist, it was all displacement maps. So to get the water running down these surfaces and look like it was realistically running down the surfaces, I developed a force-field based on the displacement textures. This meant we didn’t have to derive or get someone to model the actual geometry, we could just use the actual textures that were being used in the render. So we got water flows that looked like they were interacting with real geometry when they were actually just interacting with the texture. The result was a huge time saving for the studio as they didn’t have to generate geometry specifically for the fluid simulations. And the fluid interaction could be as detailed as the texture maps. Andy took the developed tools and tweaked them to a great look, then deployed the results into a bunch of shots very efficiently, minimizing how many unique elements had to be generated.
Asylum had already completed work on the fluids in National Treasure 2 before Fusion was asked to come aboard, what was that?
Asylum had developed a completely new methodology for getting water surfaces out of Real Flow. They were able to convert the mesh surface to displacement maps which mean you can composite different layers together to add as much detail as they wanted to a water surface wherever they wanted it. So this meant they did not have to do a really complex simulation with everything involved. Instead they could very quickly art-direct individual features of the lake movement, ripples and shore line interaction. The maps could also be easily tiled and scaled to fit the scene and added with other larger displacement maps to break up any repetition. With Displacement maps, single polygon mesh which is really light and easily manipulated in a scene could produce high resolution water surface detail at render time.
Why did Asylum ask Fusion to work with them on the fluid fx in NT 2?
Asylum was doing really well as it was, they have a great deal of experience with RealFlow and produce great work. Mainly they wanted to take advantage of the new RealFlow 4 tools to help stream line their existing pipeline to ease a tight deadline.
Why did you want Mark to do this? (question answered by Jeff Werner, head of CG at Asylum VFX)
We were really impressed with Mark’s work on Poseidon, The Guardian and other projects and we knew his experience would be invaluable on the show. Don't let the Doctorate in fluid dynamics fool you, he is very VFX savvy and knows where to draw the line between a\ perfectly accurate water simulation and something that looks great but will render in a reasonable amount of time.
Because he has such a great rapport with Next Limit, he arranged for them to visit while they were in the area. The Real Flow developers were incredibly open and receptive to ideas and issues we were having and Mark's relationship with them certainly helped. They even revisited later in the week to work technical issues with our sys Admins. The face to face interaction with developers who are based half way around the world was huge. It was a whole new level of support.
Back to Mark, what was challenging about the work you did?
Adapting a variety of Fusion’s pre-existing work flow tools to fit with Asylum’s system set-up and with how their artists work was quite a challenge.
Some of the new tool development was quite technically difficult, like the remote launching of simulations. This hadn’t been done before. It involved writing python scripts that would control RealFlow simulations on Asylum’s render farm and would communicate with their network. Because python is embedded in RealFlow, it generally works within the RealFlow instance and not with any knowledge of what’s happening in the system outside the instance.
Some of the fluid behaviors, especially the turbulent river have not been done with RealFlow for live-action shots. RealFlow is well known for not producing these kinds of flows out of the box. It can produce laminar viscous flows like chocolate or lava with ease. But to get something that looks like a turbulent, torrential river, it can’t be done by hitting the ‘sim button.’
I had to figure out a way to get turbulent style fluid motions into the particle sim to make it controllable and to make it fast. It had to be fast because to make it look realistic we had to have very large particle count sims. So I had to develop a methodology of getting dynamically moving vortex style motion into the fluid. I had a lot of experience with this kind of flow from Fusion’s fire development. Our fire sims get their turbulent behavior as a result of advanced python scripts which emulate turbulent style forces. We couldn’t apply these directly to the water because the dependence on the python scripts would force the fluid sim to be single threaded. And that would mean the river sim would run too slowly to make it possible to deliver the shots. So I adapted the methods to be able to derive vortex motions from a very low res sim and then transfer those motions as trajectories to the high res simulation allowing it to run fully multi-threaded (ie. fast!)
You seemed to have to manipulate RealFlow and write a lot of custom python scripts to make RealFlow do what you wanted it do? Why did you choose RealFlow in the first place?
RealFlow is an excellent package that provides a sophisticated simulation environment that couples fluid dynamics with rigid body dynamics in a physically accurate way. On top of that, it is fully portable between platforms and 3D packages. It doesn’t matter if a studio is using Maya or Lightwave or if they’re on Macs, Linux, or PC or a combination of all of those.
And RealFlow has python deeply embedded into its structure which allows us to take entirely unique approaches to solving particle fx issues in productions. The python scripting allows me control over the physics of the simulations, the post-processing of the simulation data, and any pipeline aspects. And now that python is becoming a ubiquitous programming language, it’s being used for render farm management, tool development and is embedded in other software packages like Maya and Houdini. We’re able to make RealFlow simulations talk to diverse elements of a studios pipeline.
What was it like working with Asylum?
It was remarkably low-stress given the intense schedule, with a staff of incredibly fun, talented and very smart artists. The team atmosphere made for a really stimulating exchange of ideas and creativity which resulted in the fast and open development of new methods. You never felt like you were struggling to be heard or competing with egos, everyone was working toward the common goal and wanted to do the absolute best job possible. And it shows. All of their work in National Treasure: Book of Secrets is outstanding.
Is this the kind of work your studio normally does -- working in-house with another vfx studio on a project?
For large features we have done this kind of thing in the past – we worked in-house with CIS Hollywood on Poseidon, building their fluid fx pipeline and developing methods for violent, photo- real, large-scale water (the first time this had ever been done with RealFlow). And for Flash Film Works on The Guardian, we worked in-house building the fluid fx pipeline, developing tools and supervising the fluid fx team. We also provided significant consultation to Luma Pictures on Primeval – optimizing and stabilizing their simulations and determining methods to get the sims to behave the way they wanted them to behave.
The needs of our clients usually determines how we work. As our work is getting better known, we’re getting more interest from vfx supervisors to complete entire fluid shots in our own studio, but most of our work involves providing fluid assets – primarily for commercial work, but also network ‘idents’ and music videos. Our clients often prefer that we provide the fluid assets because it gets all of the fluid work off their plates and then they can concentrate on the vfx they do best, and they don’t need to find & train artists or devote sections of their hardware resources to fluids. And it usually it means they don’t need to deal with r&d. At the same time, they retain creative control because we can deliver the full 3D assets rather than rendered images. This means they can they re-light, re-texture, re-time and alter camera as much as they want, long after we’ve left the picture, and they can even re-use the data for other shots or projects.
What’s next for Fusion CI Studios, you seem to have developed a real niche service that is really catching on!
We’ve really expanded the fluid/particle fx services that we can offer studios – from ‘outsourcing’ our fluid fx artists to tool & pipeline development to free bid support to fluid asset packages or rendered elements. A lot of studios want to do their own fluid fx in-house, but don’t want to spend vast resources on r&d and personnel, so we are often called on to help build up a studio’s fluid fx pipeline, quickly and efficiently.
And the cool thing about hiring our artists for a particular project is that they come with the backing of our studio. So if any problems that arise, they’re fully supported by our expertise to sort out complex issues.
Fusion has a few major projects coming up this year involving live-action features and an animated feature as well. Tsunami style waves are always in big demand, but we’re pushing toward the next big thing to keep ahead of the status quo. We’re often asked for fire, for example, and we’ve made some great strides developing fire using RealFlow.