James Cameron: ​‘Soon we’ll have AI cre­at­ing movies – and it’ll suck’

It’s easy to see how Yuk­i­to Kishiro’s ear­ly ​’90s man­ga, Bat­tle Angel Ali­ta, would appeal James Cameron. It’s the sto­ry of a diminu­tive cyborg boun­ty hunter with a human brain, on a quest to meet her mak­er. Unlike the robot apoc­a­lypse of the Ter­mi­na­tor films, this world posits an accord between humans and machines, though vio­lence and sup­pres­sion still reigns supreme. Cul­tur­al respite comes in the form of the high-speed, high-casu­al­ty death sport, Motorball.

After wran­gling with the mate­r­i­al for near­ly 20 years, Cameron hand­ed over his script to Robert Rodriguez who even­tu­al­ly direct­ed Ali­ta: Bat­tle Angel, and become a cre­ative pro­duc­er on the project. LWLies met with Cameron ahead of the Lon­don pre­mière of Ali­ta, and we dis­cussed the evo­lu­tion of the dig­i­tal image, the death of ana­logue film­mak­ing and the stress­es of hav­ing to dust his scale mod­el of the Titanic.

LWLies: Ali­ta: Bat­tle Angel is pow­ered by some extra­or­di­nary com­put­er gen­er­at­ed visu­als. By what met­ric are you judg­ing the suc­cess of these visuals?

Cameron: Robert and my pro­duc­ing part­ner John Lan­dau were most­ly respon­si­ble for the fin­ished work on the visu­al effects in Ali­ta. I’d cast my eye over things every week. I’d do a quick run-through of all the shots that were in progress. I’d give notes where I felt it was nec­es­sary. But it was real­ly up to them.

What kind of notes were you giving?

For exam­ple, the Motor­ball stuff was strug­gling, because it was the most pure CG sequence in the film. Every­thing else was more set-based – there was no set for Motor­ball, oth­er than the bleach­ers and the crowd. Every­thing on the track was 100 per cent com­put­er gen­er­at­ed. I thought it was not quite at the real­i­ty lev­els of the rest of the film. I sug­gest­ed a few small changes that I thought would punch it up. They were embraced and exe­cut­ed, and I think it helped. Now I think the sequence is spec­tac­u­lar. I didn’t give any notes or move­ment or dynam­ics or edit­ing or any­thing like that. Robert is mas­ter­ful at all that. But some of the CG stuff he’s done in the past – and he’ll read­i­ly admit this – had a whim­si­cal, slight­ly car­toony qual­i­ty to it, like in the Spy Kids movies.

You get the sense he wasn’t aim­ing for pho­to real­i­ty in those films.

I agree. So I had to ask, how do you push it that extra lit­tle bit? I’d say 95 per cent of the review work was done by Robert and John Lan­dau, to bring it home. John had been through that process on Avatar, so he knew it well. It wasn’t just me but it was my Light­storm team, so that’s me and Richie Bayn­ham over­see­ing all the Ali­ta facial ani­ma­tion, and act­ing as an inter­face with the work being done by WETA. WETA, of course, are the best in the world at facial ani­ma­tion, espe­cial­ly the pho­to real stuff. But yeah, you have to be end­less­ly crit­i­cal of visu­al effects to make sure they mea­sure up. If they look good in iso­la­tion, I guar­an­tee you they’re going to work in a flow. The eye is more for­giv­ing in a flow than it is on a shot-by-shot basis. I thew in a few notes on light­ing – punch up the edge a lit­tle more. It’ll make it gutsier.

It’s inter­est­ing to hear you talk about light­ing with ref­er­ence to dig­i­tal effects.

There’s a ten­den­cy on the part of visu­al effects lighters to make every­thing look too per­fect. The image is too well-filled, or well-mod­elled. You see every­thing. Real-world pho­tog­ra­phy isn’t like that. There are parts of the frame you don’t see. They just go black and there’s noth­ing you can do about it. So I’m say­ing smash it with a strong light from this side, and let that oth­er part go.

So it’s an attempt to emu­late a more messy reality?

It’s crit­i­cal to do that. The more CG there is in a scene – and I learned this on Avatar – the more the cam­era needs to be ground­ed in real-world physics and real-world optics. And that includes the flaws of real-world optics. Things like mul­ti-ele­ment lens flares. You throw a mul­ti-ele­ment lens flare across a scene where there’s a source light in the shot, all of a sud­den it looks like it’s pho­tog­ra­phy, even though it’s pure CG.

We played that game in Motor­ball as well. Though I can’t empha­sise enough what a great col­lab­o­ra­tion this was. The Motor­ball sequence was sto­ry­board­ed, pre-visu­alised, cap­tured, shot-designed by Robert and edit­ed by Robert, then exe­cut­ed by Light­storm and WETA dig­i­tal. So he had this real­ly pow­er­ful machine on the back end to com­plete what he set in motion. I wasn’t in that process any­where, only when shots were com­ing in close to final. I goosed them a lit­tle bit, over the edge to pho­to real­i­ty. I hap­pen to know a lot of tricks.

I believe you.

Isn’t that how it should be? As a pro­duc­er, you can be the guy that buys the rights, you can sit back and let peo­ple run with it and feel some false sense of author­ship because you had a lit­tle bit of mon­ey at the right time and maybe some good taste. Or you could be the type of pro­duc­er who puts togeth­er the financ­ing and you’re just a bank. I don’t want to be either one of those. I want to be part of the team. I want to be there to serve, not to dictate.

Robert and I had a friend­ship of 25 years that pre­ced­ed our work on this. We just agreed to be real­ly respect­ful of each other’s ter­ri­to­ry. We each brought to it what we knew best. And Robert’s a shoot­er – he’s a floor guy. He loves to be out there with the cam­era and the crew. I nev­er went to the set at all, that was his turf. Behind the scenes, we col­lab­o­rat­ed on the writ­ing. I wrote the script and he cut it down. If he want­ed to tweak some scenes, I’d throw some pages at him here or there. He’d rework them and send them back. We rewrote the end­ing a cou­ple of times.

Did you ever feel he was try­ing to make a James Cameron film?

He sor­ta says he did that, but I don’t think that’s the truth of it. We served a com­mon vision. I was chan­nelling Yuk­i­to Kishi­ro and try­ing to bash all those books into some­thing that would fit in a fea­ture film. And then Robert was chan­nelling me to some extent. But he doesn’t know where I would’ve put the cam­era. I would’ve direct­ed this film dif­fer­ent­ly. Any direc­tor would. Mak­ing a film is the artist’s per­son­al cre­ative moment. And that’s unavoid­able. I can cel­e­brate this film. If I go on the road with Avatar, I can’t say nice things about it. I can’t say I loved this movie. That’s a lit­tle cheeky. But I can say I loved this movie, as I didn’t direct it. I can feel a pride of author­ship to an extent, but I can also cel­e­brate it shamelessly.

In the film, Alita’s body is made from this anti­quat­ed tech­nol­o­gy that is still supe­ri­or to tech­nol­o­gy being used in the present day.

This is the inter­est­ing thing about this sto­ry, and it’s innate in the man­ga – it’s the sto­ry of a future. It’s not the future, it’s the future of the future. She rep­re­sents a tech­nol­o­gy that’s been lost for 300 years. It’s a myth­ic achieve­ment in tech­nol­o­gy that peo­ple can’t seem to replicate.

Is there any reli­able old tech­nol­o­gy that you use to make films now?

Inter­est­ing. There’s noth­ing we use now that I used when I start­ed. Noth­ing. There’s pho­to­chem­istry, which to me is obso­lete. Mechan­i­cal move­ment is the same. As gor­geous as the Panaflex cam­era was – and it was a beau­ti­ful machine – it’s obso­lete. Dig­i­tal cam­era tech­nol­o­gy is so advanced now that I wouldn’t ever dream of going back to one of those old cam­eras. Large-scale phys­i­cal minia­tures, too, they’re gone. The last time I worked with minia­tures was on Titan­ic. I still have the 42-foot long Titan­ic mod­el, but it’s like a Model‑T Ford. No, maybe some­thing high­er, like a Due­sen­berg or something.

The point is, we wouldn’t do it that way any more. We don’t work that way. We can’t. There’s too much flex­i­bil­i­ty with CG mod­els. And they don’t degrade. I’m always repair­ing that damn Titan­ic. It’s always get­ting bumped into, rail­ing get­ting snapped off, dust get­ting on the deck. We just do phys­i­cal mod­els, we don’t do mat­te paint­ings on glass, we don’t do pho­to­chem­istry any more. And I think movies look bet­ter than they’ve ever looked. I know Chris Nolan dis­agrees, and he thinks that shoot­ing on 65mm IMAX film is great, but you could achieve the same result dig­i­tal­ly if you chose to.

How long before the cam­era is obsolete?

I don’t know if lens­es and optics are ever going to be obso­lete. If you can shoot some­thing in the real world, by all means do so. I think it’s per­verse to recre­ate real­i­ty, dig­i­tal­ly, if you don’t have to. But on the oth­er hand, by all means let your imag­i­na­tion roam freely and use CG where you need to. I think we’re gonna have lens­es, we’re gonna have optics. We’re mov­ing toward the live-action pho­tog­ra­phy on Avatar, which we start shoot­ing in May, and I’m choos­ing my lens pack­age and doing my cam­era test­ing and so on.

I’m a firm believ­er in native 3D ver­sus post-con­ver­sion. We have devel­oped an incred­i­ble cam­era for Avatar 2 and 3. It’s based on their Cine-Alta cam­eras which will have 16 stops of dynam­ic range. It has 4 to 6K res­o­lu­tion. We’re putting two of them into a stereo­scop­ic rig with nine axes of motion con­trol, and 10-to-one zoom lens­es. And the whole thing weighs 29lbs. I can put it on my shoul­der eas­i­ly. It’s mind-boggling.

It’s hard to get my head around what that would look like.

I asked the Sony guys, you know this lit­tle sen­sor plate you’ve got on your oth­er­wise 25lb cam­era, how about you just took that off and gave it to me on the end of a cable? We spent two months nego­ti­at­ing how long the cable should be. We now have a means of hav­ing a light weight, nine-axis motion con­trol rig that I can eas­i­ly hand-hold. It’ll have a teth­er that’s about two-and-a-half metres long, that’ll run to a back­pack. Or it’ll go to some­thing near­by. But that rig is breath­ing in order to manip­u­late the stereo space. Sor­ry, I’m geek­ing out here.

Are we head­ing towards a future where com­put­ers will decide the shots for filmmakers?

I think you’ve got plen­ty of AI experts around who would say, yeah, sure. They’d take every movie every made, throw it into a mas­sive data­base, feed all that into an AI, deep learn­ing neur­al net­works will analyse why they work, and you’ll have an AI cre­ate a movie – and it’ll suck. Because the AI is not embod­ied, it’s not hav­ing the human expe­ri­ence. It’ll be like a film­mak­er who only knows oth­er movies as opposed to being human.

Peo­ple ask me all the time, ​‘What would your advice be to a young film­mak­er?’ It used to be, pick up a cam­era and start mak­ing a movie. Now my advice is, live a bit of life, then pick up a cam­era and make a film about what you know and what you’ve expe­ri­enced. Don’t go from being a super­fan in high school to film school, and come out know­ing noth­ing about life except what you’ve seen in movies. Because you don’t know shit. You’ve got noth­ing new to say.

I stand by that now. That’s the jour­ney I took. I left home when I was 18, I worked as a machin­ist, I worked as a school bus dri­ver, a school bus mechan­ic, pre­ci­sion tool guy, truck dri­ver, all kinds of stuff. Worked on auto body – what do you call it over here? I was a pan­el beat­er! Got mar­ried, had a house with a pick­et fence, and then I start­ed mak­ing films when I was in my mid-twen­ties. I don’t think I missed any­thing. It’s not that I was late com­ing out the gate. I mean, Spiel­berg, he was 19 when he start­ed, but he’s the excep­tion­al case.

Your ear­ly short film Xeno­gen­e­sis from 1978 is online and it’s still impressive.

It wasn’t a com­plete film, it was meant as a proof of con­cept reel to show that we could make a sci­ence fic­tion story.

The robot­ics ele­ment of that film is so great. You made these mechs which go into bat­tle. Does the tac­tile art of robot­ics still have a place in mod­ern film­mak­ing? Does physics mat­ter any more?

You don’t have to make a thing any more. I mean, 85 per cent of the new Avatar films is CG. But we don’t use robot­ics, we use peo­ple. It’s impor­tant to not uncou­ple the final image from hard physics. I’ve shot a lot of pho­tog­ra­phy in the real world, I’ve oper­at­ed all the dif­fer­ent types of cam­era myself. I think the more you embrace CG, the more you need to be mind­ful of what the image would’ve real­ly looked like had you shot it with a cam­era and phys­i­cal objects. Because then it becomes just an ani­mat­ed film. And I don’t mean just an ani­mat­ed film – it becomes an ani­mat­ed film uncou­pled from real­i­ty. If you’re try­ing to con­vince an audi­ence that what they’re see­ing is real, then you have to know. I’m a lit­tle con­cerned that a lot of CG artists only come up via the CG world. They’re not out there mak­ing images the hard way – with light. We’re very cog­nisant of that on the Avatar films.

The inter­sec­tion between live-action and ani­ma­tion seems so thin now. There’s a new Lion King movie which is being referred to as live action.

Per­for­mance cap­ture is not live action. It’s not key-frame ani­ma­tion. It’s human per­formed, but it doesn’t use optics at all. I live with this every day on the Avatar films. How do you cre­ate the illu­sion of pho­to real­i­ty? Which is real­ly what we’re talk­ing about. It has to be a com­pelling illusion.

That aim feels very human.

Sure, and when I say I that, what I mean is you have to embrace flaws. What hap­pens when the cam­era gets hit by some­thing, or some mud or water gets on the lens? The more you can cre­ate the illu­sion that there was a phys­i­cal cam­era there, the audi­ence, which has been con­di­tioned by 100 plus years of pho­tog­ra­phy, is going to believe the image that they see. Cin­e­ma is all about that belief.

Ali­ta: Bat­tle Angel is in cin­e­mas 6 Feb­ru­ary. Read the LWLies Rec­om­mends review.