| VFX Workflow |
[Apr. 10th, 2009|10:05 am] |
Now that it's finally released I can talk about my last project: "Watchmen". My company, Sony Pictures Imageworks, primarily did the Doctor Manhattan character. My contribution was the shaders - I didn't animate or render, but I wrote some of the programs that described the surface of Doctor Manhattan, specifically his eyes and skin and fingernails. Today a friend asked me the following question about the movie and I thought it would make a good post.
When you all put together a sequence, and then look at the motion, do you finesse and re-finesse obsessively to get human movements exactly right?
What I mean is once you have your model and skin and bones all set, do you make your best attempt at lifelike movement, then view it, then tweak over and over, or do you really work hard during the initial animation and do minor tweaks?
It was just something I was wondering while watching the film - Dr. Manhattan often looked flawless for a very long time, then he'd make some small move or open his mouth in a way that caught my eye as 'not right', then he'd go on to be great for a long time after. Seems like chasing down and fixing all the 'not quite right' movements would be infuriating.
At almost any studio a sequence starts with previsualization of a storyboard. Extremely crude characters are mocked up and simple renders are created to establish cut lengths and camera angles. This helps the directors figure out whether what they want to shoot will look good, it helps the CG Supervisor plan their approach, and it gives the editors something to cut into the working print.
Next the sequence is cut into shots and turned over to an animator who does a blocking pass. Simple, crude animations that help establish basic action and timing. How fast does Doc walk from one side of the room to the other? When does he pick up his tie? In animation dailies the action slowly gets refined and eventually approved by the animation supervisor.
While all this is happening look development TDs like me are doing basic design on the character, usually with static or simple motion tests, such as a character spinning 360 degrees or flexing his arms and looking around. Modelers, riggers, and texture painters tweak the character to improve its design and fix problems. (Riggers figure out how to rig the character - where the bones and joints are, which parts of the character flex and fold as it flexes and moves.) As the character's look starts coming together, the test shots and preliminary look get approved by the studio VFX supervisor and clients. (The director, production-side CG sup, etc.) At the same time match movers are setting up virtual cameras that track how the real cameras were moving when they shot the original footage.
Eventually, the character is approved and you're ready to go into full production. The animation passes get sent over to the technical directors who add the animation to the characters, put them into their environment, add matchmoved cameras, and start rendering actual shots. Technical directors will present "slap comps" - simple composites of the rendered images - in dailies so that the VFX supervisor can see how things are coming together. Some TDs composite their own shots and others send their rendered elements to compositors who merge the rendered elements into the original photography so that it looks good.
It's called a "production pipeline" because it works not unlike a series of tubes - the shots run through previs, matchmove, modeling, rigging, texturing, animation, rendering, and compositing before they're accepted by the client. And - to get to your question - at any point in the pipeline assets can get "kicked back". TDs or VFX Sups can take an asset in a shot that's being composited and kick it back to rendering for a new pass. Shots in rendering can get kicked back to animation to tweak something that doesn't look right, or kick it back to texturing to fix something in the map that we didn't know we'd be able to see. Shots in animation or texturing can get kicked back to modeling or rigging to fix a problem with shape or movement.
If the pipeline software is designed well, asset kick-backs actually aren't a very big problem. The animator exports a new animation pass and the TD's render scripts pick up the new publish, apply the new animation to the old model, and the VFX Sup is looking at the new animation in dailies by the next morning. It's not "infuriating" but it is tedious and, as the show producer will remind you, expensive. Every time something gets kicked back, the extra hours that someone has to spend fixing it come out of the show's budget which leaves less time and money to make things look better the first time. The mark of a good supervisor, a good crew, good tools, and good client communication is that the pipeline keeps running smoothly. Animation, models, and matchmoves that get approved stay approved. Characters that look good in neutral lighting also look good in shot lighting. Element rendering standards for one shot also work well in other shots, to keep look and workflow consistent. Every show has its share of problems and difficulties - sometimes epically so - but some shows are better than others and it can be a real joy to work with good clients on a well-run show. |
|
|
![[info]](http://l-stat.livejournal.com/img/userinfo.gif){kind=small}
sorenragsdale's journal
![[info]](http://l-stat.livejournal.com/img/syndicated.gif){kind=small}