Behind the Scenes

How We Brought President Nixon Back to Life

(And Killed the Astronauts of Apollo 11)

The Story

The contingency speech for the Apollo 11 moon landing was our inspiration.

It was written by President Richard Nixon's speechwriter, Bill Safire, who wrote it in case of any "mishaps" with the mission.

The most dangerous part of the eight-day journey was leaving the moon. Michael Collin's, the "third astronaut" on the mission, said he "wouldn't give better than even odds on a successful landing and return."

The Visuals

Deepfakes are made using a type of artificial intelligence called deep learning. It works a little like the human brain by recognizing patterns in data and "learning" from them.

We used "Video Dialogue Replacement" to train an AI model to "understand" facial movements. By using lots of visual data of human faces it "learns" the different facial parts and how they move together when people speak. It can then realistically reconstruct videos frame by frame.

To make the deepfake first we needed a "target" video.

Our new video would look exactly the same as this one, except the mouth would say the new words. All the gestures and mannerisms, including his head movements, hands shuffling papers, and eyes blinking, would remain the same.

We picked Nixon's resignation speech because it had the right serious and somber tone.

We filmed an actor performing the contingency speech.

As the AI could now understand faces, it was able to map our actor's facial motion onto that of Nixon in the resignation video.

The Audio

We also wanted a synthetic version of Nixon's voice so we needed to create an audio AI model as well.

Again, we had to feed the model loads of data. Our actor recorded hundreds of short clips from Nixon Vietnam speeches.

We then had two parallel sets of audio: the original Nixon clips and our actor reading them.

This enabled the AI to learn how to "translate" one voice into the other.

We then recorded our actor reading the contingency speech and the AI "converted" it into Nixon's voice.

The Complete Deepfake

We synced the video and the audio and that made our "complete deepfake."

Resources

Scroll Down to Continue

Behind the Scenes

How We Brought President Nixon Back to Life

(And Killed the Astronauts of Apollo 11)

The Story

The contingency speech for the Apollo 11 moon landing was our inspiration.

It was written by President Richard Nixon's speechwriter, Bill Safire, who wrote it in case of any "mishaps" with the mission.

The most dangerous part of the eight-day journey was leaving the moon. Michael Collin's, the "third astronaut" on the mission, said he "wouldn't give better than even odds on a successful landing and return."

The Visuals

Deepfakes are made using a type of artificial intelligence called deep learning. It works a little like the human brain by recognizing patterns in data and "learning" from them.

We used "Video Dialogue Replacement" to train an AI model to "understand" facial movements. By using lots of visual data of human faces it "learns" the different facial parts and how they move together when people speak. It can then realistically reconstruct videos frame by frame.

To make the deepfake first we needed a "target" video.

Our new video would look exactly the same as this one, except the mouth would say the new words. All the gestures and mannerisms, including his head movements, hands shuffling papers, and eyes blinking, would remain the same.

We picked Nixon's resignation speech because it had the right serious and somber tone.

We filmed an actor performing the contingency speech.

As the AI could now understand faces, it was able to map our actor's facial motion onto that of Nixon in the resignation video.

The Audio

We also wanted a synthetic version of Nixon's voice so we needed to create an audio AI model as well.

Again, we had to feed the model loads of data. Our actor recorded hundreds of short clips from Nixon Vietnam speeches.

We then had two parallel sets of audio: the original Nixon clips and our actor reading them.

This enabled the AI to learn how to "translate" one voice into the other.

We then recorded our actor reading the contingency speech and the AI "converted" it into Nixon's voice.

The Complete Deepfake

We synced the video and the audio and that made our "complete deepfake."

Nixon Training Clip

Actor Training Clip

Actor Speech Performance Clip

Final Nixon Speech Clip

Scroll Down to Continue