Entry 1 | derpfakes | September 18th 2018


September 18th 2018 


Firstly, thanks to everyone for supporting my work. I sincerely appreciate it and hope the community as a whole continues to grow on YouTube and beyond.

I'm not used to writing like this so give me a break for now and hopefully my manic ramblings will become more intelligible in time. This blog will serve primarily as my own commentary on deepfakes, from both a technical and political point of view. In such a unique moment as this, in a world of fake news and rapid technological and cultural change and its big-hitting commentators, my personal opinions concerning deepfaking and the surrounding issues might well mean little or nothing to the vast majority. As one of the early deepfakers to receive some mainstream attention, I have had a somewhat interesting perspective on the emerging notoriety of the technology and its impact and so I offer up some of my experiences and opinions to those who are willing to listen. Ultimately, I'm just an individual who makes videos that hopefully bring some smiles to some faces and prompt some to question how they react to controversial or suspicious information. After all, seeing is no longer believing.

derpfakes


From Then To Now


Through some strange circumstance, from my first encounter with deepfakes to the present day, I have been contacted by myriad of websites, journalists and users from across the world. My work has been seen and discussed by a range of news outlets – the New York Times, the BBC and all manner of television and radio shows from Australia to Brazil to Denmark. Given the novelty and potential impacts, the good and the bad, I am not surprised the technology received such widespread attention. Through luck or judgement I was simply one of the few in a certain place at a certain time and making somewhat more family friendly fakes than the majority of users on the original subreddit and so gained some early exposure.

It was especially interesting to me to see the different approaches and eventual articles from the journalists. Many, especially in the very early days, wrote shock pieces about deepfake pornography, potential application as a political weapon and the general downfall of mankind. These articles get hits. I get it. A lot of the time my quotes were used out of context or abandoned entirely, substituted instead with apocalyptic hyperbole, which whilst not particularly surprising to me, certainly did have a sense of irony given the writers' alleged panic at fake news. To make it clear, many if not most of the journalists who contacted me represented me and my words honestly and with integrity and I am grateful to those who did. Maybe just tone down the end of the world stuff a little...
 As time went on, more and more film and technology sites contacted me as deepfake bogeyman syndrome cleared. Less interested in quick grab clicks, these articles focussing on the positive potential of deepfakes began to take centre stage and more technical questions rather than political or moral ones began to be asked. I am certainly not an expert on the technology nor will I ever be, but it certainly interests me more than the wild and rash declarations based on unfounded assumptions that I kept hearing from freelancers and so I was happy to talk to anyone who wanted my input on the technical side

Just as the technical discussions continue to circle the deepfakes community, the legal debate also rages on. The core of the concern at the very highest levels is not so much the likes of fake pornography and certainly not of Nicolas Cage, but rather national treasure. I mean security. In fact this debate hasn't really even begun. Just last week, a letter was sent to the US Director of National Intelligence Dan Coats by three House representatives concerning the malicious use of the technology by both “foreign and domestic actors”, specifically regarding national security issues. Clearly any new technology with this potential must be thoroughly outwardly investigated by intelligence agencies but let's not kid ourselves, even if such institutions haven't had this type of technology until relatively recently (a whole different discussion), the idea of falsified video is nothing new and whether sooner or later this type of technology was certain to become widespread and cheaply available to the masses. The lack of foresight to have missed this possibility, perhaps inevitability, by the world's leading intelligence and security organisations is inconceivable. Lawmakers could be forgiven - not national intelligence groups.

Are there indeed foreign powers willing and able to create and effectively spread deepfaked video for the purposes of political disruption and propaganda? I think it's pretty safe to assume yes. What can be done to tackle these actions? Well, not a lot. Stopping the creation at source is nigh on impossible and given the prevalence and scope of social media and viral trends, it's a case of once it's online and starts to spread, you're looking at some creative interpretation of censorship laws at best, damage limitation at worst. But do people really need to be told what is and what isn't real?

The satirical news outlet The Onion has recently written an article titled 'FEMA Airdrops Emergency Cyanide Pills For Residents Stranded By Hurricane Florence'. How do we know this isn't true? Firstly I am familiar with the fact The Onion is satirical, but even if I wasn't, it doesn't take too much critical thinking to reasonably judge that FEMA has not actually been airdropping cyanide pills to the victims of a major natural disaster. Familiarity and context both at work.

In late 1938, Orson Welles narrated War of the Worlds on the Columbia Broadcasting System. In hearing of their impending doom at the hands of an alien invasion, there was mass panic as many had missed the notice of it being entirely fictional and had misinterpreted the reading as a genuine news report. Whilst an amusing story, it clearly shows the power of context. Nobody in 2018 walks out at the end of Infinity War believing half the people in the world have been wiped from existence by a giant purple alien with some fancy jewellery, yet show the same movie to an unaware and isolated individual from Elizabethan England who is entirely foreign to the idea of science fiction let alone CGI and there's a reasonable chance you'll end up watching them lose their shit. The Avengers is of course merely a fiction with some convincing visual effects to aid in the storytelling, but you know this because you are familiar with modern cinema, modern culture and modern technology and (hopefully) have the ability to use some common sense.

In a similar train of thought, there's a number of 'incoming nuclear missile' prank videos on YouTube. Some people fall for it whilst others don't. Try the same prank on somebody that has seen the videos and you're unlikely to scare them as they are already familiar with the entire charade. So would widespread awareness of deepfakes prevent them too being used as a political tool or a political weapon in a similar fashion? Probably not, but I would suggest it could stop the natural tendancies of some to immediately jump to the conclusion of an authentic work and believe whatever they see. That ability and desire to discern fact from fiction could arguably have a greater effect on interpretation of information within the wider spectrum of media.

For me, none of the above examples really address the scary aspects of deepfakes and other similar emerging technology. What if FEMA did drop cyanide pills, aliens did invade and an ICBM was headed for you? I suspect you likely wouldn't believe it. With the advancement of AI created video outpacing the detection of AI created video, we're left with the very real possibility that genuine footage is determined to be fake or of unknown authenticity and therefore 'of no value' to the viewer who cannot judge one way or the other and so gives up trying. The blend of both these scenarios then, where fact meets fiction meets contradiction is the true danger when discussing not only deepfakes, but news and information as a whole in the modern world. Not knowing who or what to believe gives way to something truly frightening. Apathy. Indifference.

So here at the end of my incoherent political ramble #1, I can sum my feelings as this – one way or the other, for better or worse, the cat is out of the bag. Banning deepfakes won't stop their production and would simply lead to those willing to break the law remaining as the sole users. Most countries around the world already have pre-existing laws preventing the vast majority of the nefarious uses for the technology and those who want to influence political outcomes or create revenge porn are not concerned with something as mundane as a ban. Clearly I am biased and unapologetically so, but in handing over the pen we lose Dostoyevsky, Dickens and Twain. Will deepfakes be used by foreign powers for political purposes? Yes. Would a lack of it end political subversion and propaganda? No. The written word is and will always be more dangerous a weapon than deepfakes. But then again, maybe you shouldn't believe everything you read.


That Damn Moustache


Since the very earliest days, people have requested I fix the unsettling footage of Henry Cavill’s facial-hair-free face in Justice League. Another Reddit user actually did already something very similar and so I left it all to one side, but as time went on people kept requesting and having already agreed to do it, I decided to finally give it a proper look.

Having audibly laughed out loud at the opening scene from Justice League, in which Henry Cavill’s Superman is seen talking to some young children and his lower face is both strangely terrifying and hilarious at the same time, I had little doubt as to which footage to use.

On paper it was always going to have issues. The relatively low quality (the scene is supposed to be filmed on a phone by one of the children) and somewhat high saturation meant the frames were not ideal, though usable, but the real struggle came with the close up of the face in the latter part of the clip. When training, most deepfakes scripts actually use either 64x64 or 128x128 pixel faces which are then upscaled to 256x256 for the final output. Therefore in a frame with a height of say 1080 pixels, a close up shot of a face causes a huge reduction in quality as it is stretched. This is largely why the face in my second Cavill video appears higher quality, as the deepfaked face only takes up a comparatively tiny portion of the frame. The obvious upside to the base footage in both of my Cavill videos is that as he is acting as both source and destination, there are far fewer problems surrounding pesky things like likeness and head shape.

Once I had completed the fake, I felt the need to add one extra clip – Christopher Reeve. To many people, myself included, Reeve epitomizes not only Superman, but the symbol of superheroism itself, both on and off the screen. I will be the first to remark on the morally grey area this takes us into. Using a deceased actor in any context comes with an almost inexhaustible list of arguments for and against. In briefly returning Reeve to his most iconic and beloved role, I hope to have found something of a balance - though I'm sure others might disagree. Be assured this work came from a place of respect.

Following on from this, whilst in this particular scene the original CGI was not the best, deepfakes isn't going to take over from the big VFX studios anytime soon and I have great respect for those effects artists. My tongue in cheek digs at the time and money spent on these visual effects are in no way meant to negate the work of the original artists and are just something of an early warning that the industry may soon be shaken up thanks to deepfakes. As I said earlier, the resolutions that deepfakes works at are tiny and when we're talking in the realms of 2k and 4k for the big screen, they just won't cut it. As it stands, deepfakes pretty much only win out on three factors and are destroyed in the rest.

The first two are time and money – clearly only one winner here, even when considering the needs of massive hardware allowance for high resolution deepfaking. The third factor is likeness. In my opinion the base likeness (assuming face and head shape etc. are close) of a deepfake wins out almost every time. But in terms of current resolutions, skin detail, texturing, subsurface scattering, customisation, meshes, eye detail, mouth detail, reflection, refraction, muscle movement, lighting, hair, camera placement, focal length, exposure and practically ever other factor, the good old fashioned VFX methods dominate and will likely continue to do so at the highest level for some time. But even in under a year, there has been marked progress with not only deepfakes but ‘AI' in general and given time, it will be right up there as a tool, not a replacement, for CGI and VFX teams.


Going Forward


I have a to do list. A long to do list. When deciding on my next fake, I try to balance the demands of the different areas of my fan base (Nic Cagers, ‘Fixed' CGI, Actor Replacements, Moustache Removals, Restoration, Technical) with the suitability and viability of footage and the implications present with whatever I create. I cannot, whether through technical or other issues, make everything that I would like to and therefore not everything the community would like me to. More often than not, deepfakes take a long time to make and I'm only one man - I do all this in my spare time. Having said that, if you do have an idea then please do let me know because so many great ideas have already been thrown my way and sooner or later I hope to realise the majority of them. Besides, Nic Cage still isn't in every movie ever...yet.


Thank you once again for all of your support.

Special shout out to eightbitgnosis for their contribution to my Patreon!


derpfakes







If you would like to help me cover some running costs then please head over to my Patreon page.
Even a dollar or two helps me out!

If you would like a personalised deepfake (SFW!) Check out my Fiverr. 





Comments

  1. How about Barack Obama as Lou Bega in MAMBO No.5? I mean it's his voice, why not put his face on it :)

    ReplyDelete
  2. how do you change a specific face when the clip has several people in? I did a test but it seems every faces put on a weird mask.

    ReplyDelete
  3. Actually the panic caused by War of Worlds broadcast has been largely overstated.
    There is not much evidence backing such claims up, there were a few cases of isolated incidents that were documented but by and large it appears that the overwhelming majority that tuned into the broadcast understood it was fictional. Not to mention the broadcast itself made this very clear from the beginning and had regularly scheduled breaks for advertisements.

    ReplyDelete
  4. Have you thought about continuing this blog?

    ReplyDelete

Post a Comment