(ominous children's music)

Well, well.

I shall bestow a gift on the child.

Hi, I'm Mike Seymour from fxguide.com for Wired,

looking at the tech of the making

of the new Disney film, Maleficent.

The film has a whole world of visual effects,

but there's a second world

of detailed visual effects shots that, until now, no one

outside the filmmakers has actually seen.

These completed shots are made, and then discarded,

just to try and produce realistic digital characters

for the film.

Matching fully digital characters,

especially their faces, is extremely difficult.

In the film, the three main flower pixies are

particularly difficult and challenging,

as the audience gets to see them both

as their normal life size, and then,

as their smaller, cuter pixie versions.

To make these pixies, the director turned

to face experts and specialists, Digital Domain.

The normal approach is to film actors' faces

on a motion-capture stage, and then retarget

those facial movements to the digital puppet

of the smaller pixie, which has, of course,

different proportions, has less wrinkles, bigger eyes.

So, if the side of the actor's mouth moves up for a smile,

then the digital puppet's mouth sides also move up.

But doing this direct retargeting from the actor,

straight through to the digital character in one jump,

makes it extremely hard to work out what's going wrong,

if something in the final clip doesn't look right.

Is the skin wrong, is the expression inaccurate?

Or was it just simply that we didn't like the take

that the team had selected from the original actress?

Digital Domain took a much longer and more complex path

to get from the actress to the final.

But in so doing, it broke out each stage separately,

so that each stage could be individually tested.

The motion-capture data was coupled

with light stage scanning.

Now, thanks to hundreds of computer-controlled LEDs,

this makes for a very accurate model of the actress,

but also gives us something called

image-based lighting.

This allows one to view the real actress

under any lighting configuration,

or really under any place that you could record,

say, natural light.

The team then makes a completely digital version

of the actress, textured and lit

to look exactly like the actress.

This test, for example, places her

in the production office's boardroom.

Only once they have produced a perfect copy,

do they then move on to animate

this digital clone of the actress,

with every single line of her dialogue from the film.

And don't forget, at this stage, all of

this high-end animation and final rendering

will never be seen by anyone outside the production,

especially not the audience.

Finally, once the digital actress performs

and looks exactly like herself,

they then discard all of this footage,

and retarget that to the smaller pixie version.

And then, of course, they re-render it, all over again.

The advantage of this approach is if the director wants

to change, for example, the proportions

of the smaller pixie, the team just has to retarget,

and then re-render.

And it had a second huge benefit,

of allowing one to examine everything, stage by stage,

as if it was a final shot, to hunt down anything

that maybe didn't look right.

For example, they could focus just on the eyes,

or maybe focus just on how blood flowed

around the actress' face as she talked.

This whole approach of producing

side-by-side animation may seem incredibly expensive

and time-consuming, but it's really what it takes

to produce a believable human face.

Don't forget, subscribe for more behind-the-scenes action.

I'm Mike Seymour, for Wired.

(soaring orchestral music)

(dragon roars)