Name: 電影中的 "去老化 "如何變得如此出色 (How de-aging in movies got so good)
Uploaded: 2025-03-11T13:37:37.000Z
Duration: 9 min 43 s

Let's take a trip back to 2006. Ian McKellen is 67 years old. Patrick Stewart, 66. Hugh

讓我們回到 2006 年。伊恩-麥凱倫 67 歲。帕特里克-斯圖爾特 66 歲。休

基礎被動語態

Jackman, 38, but that's not really relevant because in X-Men The Last Stand, these two guys don't quite look their age. Their characters were digitally de-aged in what is widely known to be the first implementation of this kind of technology. The effect is… well, I mean, you know, it looks like the first time they were using this technology. But since then, there have been many attempts to level up this illusion, with varying degrees of success.

傑克曼 38 歲，但這並不重要，因為在《X 戰警：最後一戰》中，這兩個人看起來並不像他們的年齡。眾所周知，他們的角色是通過數字技術 "去齡 "的，這也是此類技術的首次應用。效果是......嗯，我的意思是，你知道，這看起來像是他們第一次使用這種技術。但從那時起，有很多人嘗試將這種幻覺升級，並取得了不同程度的成功。

轉承詞

Because even at the highest levels of Hollywood, it is really f***ing hard to pull off.

因為即使在好萊塢最高級別的電影中，也很難做到這一點。

If you look at a person and there's something not quite right with their face, what our lizard brain is telling us is, ooh, that person is sick. I better stay away because I don't want to get sick.

如果你看著一個人，發現他的臉有些不對勁，我們的蜥蜴大腦就會告訴我們：哦，那個人生病了。我最好離他遠點，因為我不想生病。

I'm the visual effects supervisor on here.

He was one of the people tasked with making sure the filmmaking team could pull off an ambitious 53 minutes of de-aging work for Tom Hanks, Robin Wright, Paul Bettany, and

他的任務之一就是確保電影製作團隊能夠為湯姆-漢克斯、羅賓-懷特、保羅-貝坦尼和

Kelly Riley. Or perhaps we should say re-aging, as these extremely famous faces become much younger and much older than they are right now.

凱莉-萊莉或許我們應該說是再衰老，因為這些極富盛名的面孔變得比現在年輕得多，也比現在蒼老得多。

People that are critical of de-aging and face replacement, they have every right to be.

責備去老化和換臉的人，他們完全有權利這樣做。

We have it all worked out! We have it all worked out!

It's a very, very hard effect to pull off because we just, we look at human faces all day every day. It's the thing that we recognize the most.

這種效果很難實現，因為我們每天都在看人臉。這是我們最容易辨認的東西。

But Kevin and Hollywood have a new filmmaking toolkit, one that makes convincingly changing an actor's age much more achievable.

但凱文和好萊塢有了新的電影製作工具包，它讓改變演員年齡變得更有說服力。

De-aging actors is a painstaking process.

演員去老化是一個艱苦的過程。

Oftentimes we'll 3D scan an actor and build a full three-dimensional version of their head in the computer.

很多時候，我們會對演員進行 3D 掃描，然後在電腦中構建一個完整的三維演員頭像。

Then from there, you have to like build every muscle and how the skin moves when each muscle is activated and how the blood flows in and out of the skin as it moves and where the peach fuzz lies.

然後，你還必須鍛鍊每塊肌肉，瞭解每塊肌肉被激活時皮膚是如何運動的，皮膚運動時血液是如何流入流出的，以及 "桃花 "在哪裡。

You're trying to recreate reality from the inside out in that approach and it can yield really impressive results.

你試圖用這種方法由內而外地再現現實，它能產生令人印象深刻的效果。

The problem is, it is so labor intensive.

And since all this meticulous face swapping takes place after filming is wrapped, there's a huge amount of subjectivity and kind of filtering in between what the actor does and what ends up on the screen.

由於所有這些細緻的換臉工作都是在拍攝結束後進行的，是以演員的表演和最終呈現在銀幕上的效果之間存在著巨大的主觀性和過濾性。

Which is one of the reasons they felt that a face swap in post wasn't going to cut it for here.

這也是他們認為在後期換臉並不合適的原因之一。

It was clear that traditional means of de-aging was never going to work at the quantity that we had to address and certainly not at the quality consistently throughout the production that we needed.

很明顯，傳統的去老化方法無法滿足我們所需的數量，當然也無法在整個生產過程中始終保持我們所需的品質。

And it wasn't until we saw the results of the very first test by this company, Metaphysic, that we kind of went, oh gosh, thank, you know, we can do this.

直到我們看到 Metaphysic 公司首次測試的結果，我們才意識到，天哪，謝謝你，你知道，我們可以做到這一點。

Metaphysic enabled Elvis to perform on America's Got Talent and also made the backup singers look exactly like the judges.

Metaphysic 讓貓王得以在《美國達人秀》上表演，還讓伴唱歌手與評審長得一模一樣。

They helped Eminem dance with his younger self and brought Ian Holm back from the dead.

他們幫助阿姆與年輕時的自己共舞，讓伊恩-霍爾姆起死回生。

They're an AI company that's been making waves in Hollywood during a time where there's still some uncertainty and discomfort over the technology.

他們是一家人工智能公司，在好萊塢掀起了一股人工智能浪潮，而在這個時代，人們對這項技術還存在一些不確定性和不適感。

They use sophisticated machine learning to generate faces, but maybe it's better to let

他們使用複雜的機器學習來生成人臉，但也許最好讓

We want Tom to be 20 and then 25 and then 30 and then 35.

我們希望湯姆 20 歲，然後 25 歲，然後 30 歲，然後 35 歲。

So we kind of create these groupings of images that we can feed the AI with and teach it in the same way that if I gave you a book that thick, you know, a thousand pages of images of Tom Hanks from all different kinds of places, they might be family photos, they might be publicity stills, movie posters, images from the films he's been in.

是以，我們創建了這些圖片組，我們可以向人工智能提供這些圖片，並以同樣的方式進行教學，如果我給你一本厚厚的書，你知道，裡面有一千頁湯姆-漢克斯在不同地方拍攝的圖片，它們可能是家庭照片，可能是宣傳劇照、電影海報，也可能是他出演過的電影中的圖片。

And I said, all right, Ed, you're going to do nothing for the whole week but flip through the pages of that book and just study Tom Hanks' face.

我說，好吧，艾德，你這一週什麼都別做，就翻翻那本書，看看湯姆-漢克斯的臉。

If at the end of that week, as long as you're a visual person, if I said to you, OK, now close your eyes and imagine Tom Hanks singing happy birthday to you in the moonlight.

如果在那一週結束時，只要你是一個喜歡視覺的人，如果我對你說，好了，現在閉上眼睛，想象湯姆-漢克斯在月光下為你唱生日快樂歌。

Even though none of those images had shown that, you would still be able to do that very clearly in your head.

即使這些影像都沒有顯示出這一點，你仍然能夠在腦海中非常清晰地做到這一點。

People confuse it for like, you know, you feed it 10,000 images and it finds the closest one and it kind of collages it onto a person's face.

人們會把它混淆，比如，你給它輸入 1 萬張圖片，它會找出最接近的一張，然後拼貼到人的臉上。

It's actually, it's not using a single pixel from any of the data that it's trained on.

實際上，它沒有使用訓練數據中的任何一個像素。

So it's more like it's drawing an entirely new face with every frame as opposed to copying and pasting pieces of other photos onto the same face to assemble a new one.

是以，相對於將其他照片的片段複製並粘貼到同一張面孔上以組合出一張新的面孔，它更像是在為每一幀照片繪製一張全新的面孔。

You won't find two faces that are alike in any of the training material and what ended up in the movie.

在任何培訓材料和電影中，你都找不到兩張相同的臉。

This is what machine learning is good at, absorbing a bunch of data, in this case people's faces, and generating something new.

這正是機器學習的強項，它能吸收大量數據（在本例中是人臉）並生成新的東西。

But the particularly groundbreaking aspect of this implementation was the speed at which it could work, which changed the production entirely.

但是，這種實現方式尤其具有突破性的一點是它的工作速度，這完全改變了生產方式。

It's like really fast, like it will do its thing in like, you know, milliseconds.

它的速度非常快，只需幾毫秒就能完成任務。

That is what we use to do the live on set face swaps.

我們就是用它來進行現場換臉的。

We would have one monitor that had young Tom Hanks on it and one that had him at current day, and that was really powerful.

我們會有一個顯示器顯示年輕時的湯姆-漢克斯，另一個顯示器顯示他現在的樣子，這真的很震撼。

It allowed for immediate on set course corrections by the actors themselves because

這使得演員自己可以在現場立即修正路線，因為

Every time after Bob Zemeckis called cut, they would look and kind of go like, yeah, you know, I'm, I'm shuffling a little bit here.

每次在鮑勃-澤米吉斯宣佈剪輯之後，他們都會看一下，然後說："是的，你知道，我，我在這裡做了一點調整。

Maybe I need to stand up more straight to sell the sort of like physicality illusion or ooh, I was overacting my youth in that take.

也許我需要站得更直一些，才能讓人產生一種肢體上的錯覺，或者說，我在那個鏡頭裡表現得太年輕了。

So it really brought the actor into the conversation and allowed them to use it as a tool.

是以，它真正將演員帶入了對話中，讓他們將其作為一種工具。

It also helped them seer away from a situation where an actor has been made to look young, but their physicality doesn't feel young.

這也有助於他們避免出現這樣的情況：演員被塑造得很年輕，但他們的體態卻讓人感覺不年輕。

Like De Niro's character here in The Irishman.

就像《愛爾蘭人》中德尼羅的角色一樣。

He's supposed to be a rugged and tough mobster here and I mean, he just looks like my grandfather trying to kick someone while they're down.

在這裡，他應該是一個粗獷、強悍的黑幫老大，而我的意思是，他看起來就像我的祖父，試圖在別人倒下的時候踢他們一腳。

Shout out to this guy for selling it though.

And I mean no disrespect to the absolute powerhouse of an actor that Robert De Niro is.

我並不是對羅伯特-德尼羅這位實力派演員不敬。

Obviously, this kind of thing just couldn't be addressed because the de-aging was happening solely at the end of the pipeline.

顯然，這種問題無法解決，因為去老化僅僅發生在管道的末端。

It's very easy to look at a 67 year old acting young and think, God, they look so young.

看著 67 歲的老人裝嫩，很容易讓人聯想到 "天哪，他們看起來真年輕"。

But then if you were to put a 20 year old face on them, they would look like they were a 20 year old's face on a 40 year old body.

但是，如果你把一張 20 歲的臉貼在他們身上，他們看起來就像是把 20 歲的臉貼在了 40 歲的身體上。

And it's just like very lizard brain reaction.

It's something deep inside of us that tells us that something isn't quite right.

我們內心深處的某種東西告訴我們，有些事情不太對勁。

If you look closely, you can see that the real-time de-aging works on the face, but it leaves the ears and the neck untouched.

如果你仔細觀察，就會發現實時抗衰老對臉部起作用，但對耳朵和頸部卻沒有作用。

That didn't really matter because the other benefits of the real-time feed was that the editors could use the half-baked footage to assemble early drafts of the film.

這其實並不重要，因為實時傳輸的另一個好處是，剪輯師可以利用這些半生不熟的片段來組裝電影的初稿。

It was incredible walking away from set with like every shot that was going to be a face swap shot.

從片場走出來時，我覺得每個鏡頭都會是換臉的鏡頭，這太不可思議了。

There was already a very rudimentary version, but pretty convincing version of it already done.

當時已經有了一個非常簡陋的版本，但相當令人信服。

And then in post-production, we used very similar AI tools, but much, much higher fidelity.

在後期製作中，我們使用了非常類似的人工智能工具，但保真度要高得多。

So instead of doing, you know, a face swap in 10 milliseconds, it'll take a few minutes, but that way we can get like the real film quality, high fidelity results out of it.

這樣，我們就不用在 10 毫秒內完成面部交換，而是需要幾分鐘時間，但這樣我們就能獲得真正的電影品質和高保真效果。

When it comes to making people older, I had to ask Kevin about this scene because I found it so moving.

說到讓人變老，我不得不向凱文詢問這一幕，因為我覺得它太感人了。

And the performances are so good, especially Paul Bettany.

表演也非常出色，尤其是保羅-貝坦尼。

Rose, she, she loved having you all around her.

羅斯，她，她喜歡你在她身邊。

I was like, oh my God, it just, I really felt it.

我當時想，天哪，我真的感覺到了。

And I imagine that's one of the scenes where you guys had to age him up a little bit.

我想這也是你們必須讓他老一點的場景之一。

That scene actually was almost purely prosthetics and makeup.

那場戲實際上幾乎完全是假肢和化妝。

But with other scenes in the film, they were able to marry the physical prosthetics and digital makeup to create something that wouldn't have been possible with either alone.

但在影片的其他場景中，他們能夠將實體假肢和數字化妝結合起來，創造出單靠其中任何一種都無法實現的效果。

Robin has never been old before, so we couldn't train these models on family photos of her at 85.

羅賓從來沒有老過，所以我們無法根據她 85 歲時的家庭照片來訓練這些模特。

But what we could do is we could use other AI driven tools and workflows to feed images of her from set and say, Hey, give me a version of this with better skin translucency.

但我們可以做的是，我們可以使用其他人工智能驅動的工具和工作流程，從片場輸入她的影像，然後說，嘿，給我一個皮膚透光度更好的版本。

Or one that wrinkles around the nose in a way that feels more like stretching skin and less like a flexible prosthetic.

或者在鼻子周圍起皺，感覺更像是拉伸皮膚，而不像是靈活的假體。

And so what we ended up doing is employing the same tools that we use for de-aging and helping them for up-aging.

是以，我們最終所做的，就是採用與去老化相同的工具，幫助他們去老化。

And this mixing of technology is what helped sell so much of the film.

而這種技術的融合正是這部影片的賣點所在。

People kind of talk about AI dehumanizing certain things.

人們常說，人工智能會讓某些事情失去人性。

And in this case, I think it really humanized the whole process because it took these people with decades of experience and allowed them to use it in order to better the film.

在這種情況下，我認為這真的使整個過程人性化了，因為它把這些擁有幾十年經驗的人帶到了電影中，讓他們利用這些經驗來更好地完成電影。

What we're doing now, it's cutting edge in the future.

我們現在所做的，在未來將是最前沿的。

Hey, if you made it this far, thank you so much for watching.

嘿，如果你能看到這裡，非常感謝你的收看。

Videos like this, deep dives into industry trends and movie technology that have really deep research and reporting, but are still accessible to watch are some of my favorite to make.

像這樣深入研究行業趨勢和電影技術的視頻，既有深入的研究和報道，又易於觀看，是我最喜歡製作的。

Partly because it's something that I wish that I saw more of.

部分原因是我希望能看到更多這樣的事情。

I watch a lot of breakdowns and deep dive videos geared towards nerds like me that care so much about the tiniest little details of cinematography or animation.

我看了很多針對像我這樣的書呆子的分解和深度挖掘視頻，他們非常關心電影或動畫的最微小的細節。

But here at Vox, we want to make great independent journalism that's accessible for anyone, whether you're an industry expert or just somebody who's a little bit curious.

但在 Vox，我們想做的是任何人都能訪問的優秀獨立新聞，無論你是行業專家還是隻是有點好奇的人。

And that's something that we're only able to do thanks to the support of our members.

我們能做到這一點，全靠會員們的支持。

So if you like that mission and want to ensure that videos like this continue to be made on the internet, you can support us at vox.com slash memberships.

是以，如果你喜歡我們的使命，並希望確保這樣的視頻能繼續在互聯網上傳播，你可以通過 vox.com slash memberships 支持我們。

So thanks to everyone who's already supported and thanks again for watching.

感謝每一位支持者，再次感謝您的收看。