How ILM Built a Digital NYC for Marvel's 'Avengers' (Just to Blow It Up with VFX)
ILM has been at the cutting edge of visual effects in cinema since George Lucas founded it back in 1975. Constructing some of Hollywood’s most memorable moments is an impressive feat, so it’s only natural that Joss Whedon and company tapped the FX house for the much-anticipated The Avengers. Marvel’s recent efforts in film (Iron Man, Thor) built quite the lead-up for Avengers to deliver upon, and needless to say, the film brought the spectacle in spades. Check below to see how ILM built a massive digital ‘stunt’ version of New York City, just to rough a whole bunch of it up — plus some more from ILM on other VFX in The Avengers.
This process sounds like an incredible headache, but the attention to detail really creates a striking effect. I particularly appreciated the projection of the offices to the building’s interior. Of course we have to see into the building! Even flying by at high-speed, real geometry just sells the depth better than faking it would. This isn’t the only VFX work ILM crafted for Avengers, either. Here’s more material straight from the company’s official YouTube channel (there’s some redundant information here, but plenty new info overall):
Even though I’m one of those negative nancies that believes CG stunt doubles will always look ever so slightly ‘off’ no matter how skillful the artists and animators, the Hulk of The Avengers holds up extremely well. Even (or especially?) under dramatic lighting, there’s palpable texture and muscle there, and a lot of expressiveness that seals the deal for realism. Because of its mostly static nature, a city is a lot more straightforward to recreate in CGI than preserving an actor’s performance with a digital character. Also, it seems ILM has done so in a way that looks a bit more low-impact for the actor than the ‘helmet-cams’ used for Avatar:
What do you guys think? Do you buy the ‘stunt’ CG city? Does it all hold up to your eyes? Have you ever heard of using pana-spheres or similar methodology on such a large scale?