1 INTRODUCTION

I am a random person on the internet who would like to be anonymous. I have no ethos and am practically nothing more than a concerned citizen.

I dug deeply into the speech where Donald Trump conceded the presidency, released on January 7th, 2021, and what I found disturbed me. Please read keep reading if you desire to live in a democracy. This kind of thing is coming for all of us.

One day, I was on Twitter and I saw this.

It seriously fascinated me, and I was inspired to dissect the video to crack open what secrets I could. Here is the copy I downloaded. It comes from The Hill YouTube channel, boasting a 1080p resolution and 30 fps.

Presently, I will share my findings with you, but I must first furnish context. Upon sharing my findings with close friends, I learned that the general public knows a lot less about deepfakes than I thought. In case you really just can't believe that this speech is a deepfake, please skip over to the section

then come back here and read what you passed over. If that isn't the case, then just read everything in order and let's get the important stuff over with.

If you want to confirm anything for yourself, please download this video, a copy of the speech with an overlaid timestamp, and zoom in with editing software.

2 SAME FRAMES

1080p downloadable version

What exactly are we looking at? Well, I combined 2 clips, which are 1 minute and 20 seconds apart from each other, and layered one on top of the other at 50% opacity. In a single frame, there is as much of one video as there is the other. Let's zoom in.

When the video is blurry, that means both clips are different. When they get sharp, that means both clips are simillar. What is unusual is that, for roughly 19 frames, they are crystal clear.

The original, zoomed out video is 1080p, and the pixels are nearly on top of each other. In the real world, this translates to millimeters of distance. Donald Trump would have had to place his head, and simultaneously make a facial expression, in practically the exact same way.

His head had to be in the same position and rotation within millimeters of how it was 1 minute and 20 seconds previously. Simultaneously, his facial expression and movements had to match within millimeters of how it appeared 1 minute and 20 seconds previously.

If this is a real recording of Donald Trump, then all of this would have to be the case while he spoke completely different words. Let that sink in. He was not saying the same exact thing in the same exact way, he said 2 completely different things.

Some people have told me that this is the result of video compression. Let's forget the fact that this type of video compression doesn't seem to exist. This video compression, where non-consecutive frames far away from each other are made to look simillar is not something that I have been able to track down.

But let's run along with the idea that this is some secret only the U.S. government knows about. Why would they use it on a speech of the president? Why would they compress down the speech so much that portions of some of the frames are deleted entirely and replaced with portions from other frames?

Removing frames. That's pretty much desecration, I have never heard of such a thing in regards to maintaining the recordings of speeches of important world leaders.

In some 3D cartoons, especially on rather inexpensive productions, very similar frames appear relatively often. A "3d keyframe" is sculpted for a talking character, let's say it's the character mouthing an O. A second one is sculpted later, let's say the mouth is closed. The frames in between are generated as estimations of what happened. One of these keyframes, the O keyframe, may be copied to be used again while the character is saying a completely different thing.

A deepfake can run on the same principle. There could be a serious time crunch, and rather than have a person or computer generate a keyframe for that exact moment, the same keyframe is used. I believe that this is what we're seeing here, because it is pretty much impossible for Trump to have performed this in real life, on accident, with millimeters of accuracy, saying 2 absoultely different things. Actually, let's look frame by frame.

I had a very difficult time coming up with which configuration of both videos matches the best. Here are 3 other attempts I made:

1080p downloadable version

1080p downloadable version

1080p downloadable version

You're probably still not be convinced. There's more.

3 SPEAK FRENCH TO ME

1080p downloadable version

I don't think I have ever seen anyone do this sort of thing with their mouth. I definitely don't think I've ever seen Trump do this. The lips on the left side of his face (his left, not your left), curling like that. I'm not sure if I have to go into detail.

When I first saw it I thought maybe I was just overthinking. I spent a lot of time, looking in the mirror, trying to recreate this mouth movement that Donald Trump supposedly made. Try it yourself.

Let's see the individual frames.

It begins looking normal, quickly devolves into unnatural, and goes back to normal. I see the entire thing as serious proof of the video being a deepfake. I believe that this oddity is the victim of one or two bad or missed keyframes meant to carry the lips from being closed to fully open. Another theory is that the teeth element of the deepfake made a glitch, and the grey area in the top corner of his mouth is meant to be his molars at the wrong perspective.

Some of those who I showed this to believed that this was yet another example of compression. Clumping, where new shapes are created due to compression, is something that happens, but the bitrate, which is the amount of data in every second of the video, must be very low. Right away I could tell that this was likely not the case. To show why, I have exported the speech video at a bitrate just low enough to cause clumping.

Compare that to this

As you can see, clumping destroys a lot of detail. In the uncompressed frame, there is clearly a pale rim on the bottom right corner of his mouth and a grey rim on the top right corner. You can see mildly complicated shapes. This sort of detail does not appear in the "clumped areas" of the compressed frame. Things are blocky. Clumped areas exist as regions where color and shape is simplified.

But let's say this is yet another example of secret government compression tech. For that rebuttal, I ask this: how does this detailed distortion compress the video? Consider what these frames would look like with a normal mouth, and ask yourself how having these detailed shapes jutting into it ultimately removes information. I can't see how this possibly would equate to less data in the video, and I earnestly welcome a rebuttal from a video compression expert.

4 WHIPLASH

I've heard that's a good movie. I haven't seen it.

Throughout the video, Trump appears to move his neck far too fast, or otherwise in an unnatural way. I realize that this is a subjective matter, nonetheless I had seen plenty of other people call his neck movement in this specific speech "uncanny" prior to my examinination of the speech in a video editor. Furthermore, I personally know a few people who were unconvinced by the repeated frames and mouth distortion, but capitulated upon seeing these 4 sets of clips.

1080p downloadable version

1080p downloadable version

1080p downloadable version

1080p downloadable version

When some people see these clips, they immediately sense something offputting. Others have to stare a while. Some just don't see anything. It's a subjective matter, and the next section is going to discuss this subjectivity.

5 THE POWER OF DEEPFAKES IN 2021

All rights reserved. I am displaying this for educational purposes, namely educating people on deepfakes.

This clip is from a video uploaded on April 17th, 2018 to the BuzzFeed YouTube channel. According to BuzzFeed's own article discussing it, it is the result of a free to download program, called FakeApp, after it ran for "roughly 56 hours" while being overseen by a single "video effects professional."

This was the state of deepfake technology in April, 2018. From a distance and with the volume off, it is indistinguishable from reality until Jordan Peele shows up on the side. And, it was accomplished with a free to obtain tool and one man at the computer.

With that said, the illusion falls apart once you've zoomed in and slowed it down.

The technique of zooming in and slowing down is not something I came up with. Neither is it foolproof. However, trusted computer science sources reccomend it as a method for identifying deepfakes.

Speaking in Buzzfeed's article on the video, renown computer scientist Matthias Niessner said "Slowing a video down so that you can clearly see the frame-to-frame transition can sometimes reveal temporal glitches that are introduced in manipulated video."

Right now I'd like to stop and apologize to Mr. Niessner, should this article reach him, for using his quote. I understand that you will inevitably say that you don't endorse the idea that Trump's January 7th speech is a deepfake. I genuinely sympathize and mean no offense when I say that one's reputation is important.

Cybersecurity firm Norton, in an article on its company website, explicitly defines "Images that look unnatural when slowed down," and encourages using "video-editing software that can slow down a video's playback." It also advises to "zoom in and examine images more closely."

Returning from that tangent back to the Obama video, perhaps the easiest way to describe why it doesn't look real is to say that "it looks like a videogame". The second easiest way is perhaps "it looks like a moving painting". It looks like a 2 dimensional image being distorted, stretched, skewed. Only in this zoomed in and slowed down state does it cease to appear realistic. These are ideas I want you to be thinking about when identifying deepfakes.

Now I'd like to jump forward 2 years.

All rights reserved. I am displaying this for educational purposes, namely educating people on deepfakes.

This second clip is from a commercial uploaded on September 10th, 2020, and far beyond the caliber of the previous video. Despite this, it also was produced by a relatively small amount of human resources.

According to this article from AdWeek, it was produced by only one agency and one production company.

The production company, Tribbo, employs only 28 people.

The agency, Big Family Table, employs only 13 people. The only video on their website is this same ad they made for Hulu, despite it being 4 months after its production.

Two companies, each with less than 30 employees, and one which appears to have relatively little on its record, were able to produce this incredibly lifelike deepfake.

Now lets zoom in and slow it down.

This time, the oddities are much less noticeable. It takes a little time to realize the first guy looks strange. You need to stare for a bit longer at the second guy, and it wouldn't suprise me if some people require at least a minute to notice his inconsistencies.

This was the state of deepfake technology in September of 2020. One production company and one agency, each with less than 30 employees, could produce something that, even when you are zoomed in and slowing it down, there is some difficulty in identifying it as fake. Admittedly, compared to the Obama video, many more people were involved in the creation of this ad. However, also keep in mind that all I showed you is a small clip. The complete commercial is much more ambitious than the Obama video.

Additional manpower is also the result of there being 8, not just one, different shots featuring a deepfaked face. Also note that some shots included a deepfaked face on a moving actor, occasionally making 90 degree head turns which apply highlights and shadows.

Once again, this was accomplished by one agency and one production company, each with less than 30 employees, 4 months ago.

Credit for the image goes to Parker Higgins, it is in the public domain.

If you looked into a running digital camera that was connected to the internet at any point between 2007 and right now, the NSA has video captured your face. That's a fact. The project is called PRISM.

The UDC (Utah Data Center), contains audio and video from every human being who has looked into such a camera since 2007, among many other things. And not just people in the US. Every person in the world who has lived in a country connected to the global internet. Living in any country short of those like North Korea. The UDC contains at least 5 zettabytes of data; I say at least because it was 5 zettabytes in 2013 and there have been no updates on how much it stores since then. More ambiguous is the Hawaii Cryptologic Center in Oahu, and High Performance Computing Center-2 in Fort Meade, Maryland. Everything I have said thus far is not new, nor should it suprise you. None of this is classified, none of this is new, none of this is a conspiracy theory. All of this is widely accepted, and has been reported on by mainstream news sources for several years at this point.

CNN

The Guardian

NPR

National Geographic

CNET

Wired

New York Times

Washington Post

PBS

NSA Press Release

GCN

Informa, PLC

We don't know what exactly the NSA is keeping this data for. This file transfer information, these text messages, these emails, these video captures.

Consider what the front facing camera on your phone, when it's on, is most often looking at. Your face. If your laptop has a camera, when it's on, what is it most often looking at? Your face. How often are these things looking at your face?

Consider something else. When I suggest Trump's January 7th concession speech is a deepfake made by the US federal government, I am not talking about something that was made with old hardware. I am not talking about something that was made with old software. I am not talking about something that was made by one small company. I am not talking about something made by two small companies.

I am talking about something that was made by the single most powerful government which has ever existed in history. I am talking about a massive syndicate that is backed by the biggest defense companies in the world, gets help from the best colleges in the world, and uses the definition of cutting edge technology. To put it simply, the deep state is where cutting edge technology is created.

At this point, would it really be a shock that the NSA's unspeakable quantity of video has been used for the most advanced deepfakes in the world? Would it really be a shock that these videos have been analyzed in order to create the most incredible replications of the human face ever?

The NSA has been receiving data from every computer, every camera and every microphone that is connected to the internet since 2007. I can't tell you exactly why they would have made a deepfake of the president saying these things. But consider that they are absolutely capable of producing a deepfake more advanced than anything most people, let alone most contemporary deepfake experts, are used to.

And, after 13 and a half years, having collected all of this video and audio data from billions of human beings, why wouldn't they use it to research how the human face moves? Why wouldn't they apply it to things like deepfakes? These people have every resource to produce the greatest deepfakes ever.

Now, they have come up with a recreation of the president of the united states making a speech. And this recreation is so perfect that the only reason I know it's a deepfake is, not because of the entire sequence, but because I have payed close attention to various key moments sprawled throughout.

If this was your starting place because you're skeptical, please head over to the section

to see the real stuff. If you were more open minded and read through everything in the order it's been lined up, please continue.

6 SLIPPY SLOPPY

This section is meant for people who aren't in a hurry. If you are in a hurry, this most likely won't convince you of anything and you might as well skip to section 7. When you have the time, please follow through this. Originally, my final piece of video analysis was to showcase various parts of the speech where I thought it looked the most like a deepfake. The parts where things "slip". I would ask you to consider the things you saw in the above deepfake examples and observe the same sorts of things here.

Eventually, I realized that this would be too reliant on subjectivity. Some peope just won't see anything weird in this. Others might require several minutes before noticing something like an ear moving independently of the rest of the head.

This is why I have instead decided to simply place the clip here, zoomed in and in 0.25x speed. Having memorized and considered the visual intricacies, having seen what makes a deepfake look fake, and knowing the context of how advanced the tools are which the federal government has at its disposal, simply take a look and try to find whatever strange things you can. The fact is, most of this video lacks any such signs because, if it is a deepfake, and the government does own the resources to make a deepfake of this standard, then it would be the most advanced deepfake ever.

I want you to be as convinced as I am.

7 F ON YOUR REPORT CARD

This section has since become useless. I already realize that this article will suffer intense castigation. Every side will label me something undesirable. Despite this, I still think it's necessary to discuss the analysis of actual deepfake detection software. This deepfake detection website, at one point, was capable of accepting videos. As of January 14th, 2021, or possibly even earlier than that, that is not the case. This malfunction seems to be abrupt, there is no notice when things will get back online, and my attempts to contact have been fruitless. As a result, you can not replicate the results shown in this video, which performs 2 tests.

Deepfakes are always based on real images and videos. This software detects which frames are most likely real, and which are most likely fake, e.g. were generated by the computer to fill gaps between the real ones. A friend of mine performed the second test and recieved the same result derived in the above video.

Unfortunately, I didn't record my result back when the website was still operational. I also received "it's fake" when inputting Trump's January 7th concession speech and "it's real" when inputting other Trump speeches, more specifically, his November 3rd, December 9th and December 22nd speeches.

Once again, I absolutely understand dismissing this section of my article. I have given you no way of replicating the data I've shown in this section. In other sections, I have provided a way to replicate the data that I delivered. Reminder: I encourage anyone with editing software to download the timestamped copy of the speech (or download the speech and timestamp it yourself) then go to the clips I have displayed.

While I expect you to disregard this section, nevertheless I don't think I could have adequately presented my case without mentioning deepfake detection software.

8 CONCLUSION

I have no idea why the federal government made and released this video. I have no idea if and how Donald Trump authorized this. I have no idea what any of this entails for Trump.

All I know is that this is a deepfake.

If you've read this far and I've made you seriously consider the possibility that this is a deepfake, please, archive all of the videos I have featured here, and their links. Archive this webpage as an html file, this can easily be done in Chrome and Firefox, even on a phone. Note that saving an html file saves neither the videos nor the images. Share this with as many people as you can. At this moment, at least 29% of the country does not believe Joe Biden is the president-elect. Whether or not that is the case, Donald Trump is still the President of the United States. And the people need to know that this speech, released on January 7th and supposedly from the President of the United States, was completley fabricated. It was neither spoken by, nor filmed on Donald Trump, yet it has been passed off as real with zero mainstream news support for the idea that it isn't.