[SOLVED] : “Deepfakes”: how are hyper-rigged videos made? Part three
How to create a DeepFake?
To create a DeepFake, you need a rather powerful computer and arm yourself with patience. Then, you let the artificial intelligence calculate the different video images of your special effects. One main application is available on the market: FaceSwap.
FaceSwap is a program written in Python that will run on multiple operating systems including Windows, Linux, and MacOS. It is free and fairly easy to use. Its purpose is toFace swap on a video.
For use FakeSwap, the user must feed it with hundreds of images of a character A (the target) and a character B (the one whose face will be applied). One possible source is Google, DuckDuckGo, or Bing. Videos (taken from interviews, public speeches, or movies) are a better source of images, as they will capture many more natural poses and expressions.
The program will then put in competition these bases A and B thanks to the neuronal learning, until a convincing result is obtained (the software indicates that an error rate of 0.02% is acceptable). Then just press “Create” to start the production of the faked video. For more info, see their official site.
The danger of DeepFake
One can imagine many uses of DeppFake, here are some of its uses.
Fake news or manipulation of information
At a time when “false informationSeem to contribute to most of the world’s woes, deepfakes have taken on the role of a propaganda tool. This video from Barack Obama show me how deepfake can be used for manipulate the truth and disclose information, en masse.
Deepfake offers a very powerful mechanism for those who would like manipulate public opinion.
P *** o involuntary
At present, most deepfakes are used in the‘industry du p *** o. A recent study on this practice found that women constitute the majority of victims in the p *** o deepfake. In these cases, their faces are inserted into existing pornographic content. The goal is to ruin a person’s reputation targeted beforehand.
Deepfake is the ideal technology to make improve social engineering used by hackers. Their strategy? Audio files reproducing your boss’s voice are generated using artificial intelligence. They make a call and order you to make a money transfer. And presto, voila! This has already been done with the case of a British CEO who was tricked into transferring $ 240,000 to a fraudster.
Social engineering, a tactic used in many cybercrimes that manipulates natural human behavior by playing with our confidence, urgency, shame, etc.
The sextortion, for example, is theperfect deepfake app. Sextortion is a form of blackmail. It’s when someone threatens you to send an intimate photo or video of you to other people if you refuse to send them money.
The scam usually takes the form of an email informing you that you have been caught on camera, in a “compromising position”. To prevent the video from being shown, you have to pay a ransom in bitcoin.
Fake videos or fake audio files are getting more and more realistic and easier to make. The best weapon for this kind of rigging is your vigilance. So stay vigilant, scammers will always be one step ahead!
How do deepfakes work?
Digital – Les deepfakes, or hypertrucage, is an image synthesis technique based on artificial intelligence. The term deepfake is a portmanteau word formed from deep learning and fake.
It is used to superimpose audio and video files existing on other videos existing on other videos (for example: the change of face of a person on a video). These fake videos, often funny, are sometimes taken seriously by Internet users, wrongly.
This technique can be used to create malicious infox and hoaxes and is also notoriously used to create images of people without their knowledge.
Deepfake videos multiply on Facebook, Tiktok and sometimes sow doubt. The reactions have been numerous on social networks in recent days, because this technology asks several questions.
Deepfake: How easy is it to create believable fake videos?
Creating surprisingly realistic fake videos should get easier and easier. Both in terms of image and audio.
A Barack Obama who calls his successor a “complete jerk”. A Donald Trump calling on the Belgians to withdraw from the Paris agreement. Surprisingly realistic deepfakes appeared last year. Are these faked videos which often play at substituting one face for another complex to make? Not especially, explains Vincent Nozick, professor and researcher at the Gaspard Monge computer lab (LIGM) who is working on a Deepfake detection program called Mesonet: “You just need to have computer basics to create this type of video. The algorithm is available on the Internet. ”
The first step is to build a base of photos. “You just need to collect around 10,000 images of the person you want to imitate. The ideal is to have lots of small videos where it appears from various angles of view and lighting ”, specifies the researcher. A base in other words easy to build if it is a question of making a substitution between two very media people (politicians, movie stars, etc.)
Substituting the face of an accomplice “actor” for that of a personality is hardly more difficult, however. “Ideally, you have to take a person who looks a little like him and film him from different angles and lighting, asking him to do various facial expressions,” explains Vincent Nozick. Videos that do not require more than a day of filming and which, once in the box, can serve as the basis for many Deepfakes. No need to redo this work each time: once the base relating to the accomplice is created, it is possible to rely on it for all the videos.
Five minutes of target video
“Another technology, Face2Face, makes it possible to produce faked videos with even less data,” explains the LIGM researcher. It is very complex to code and for the moment its code is not accessible. It should therefore remain “secure” for a few more years, but eventually others will manage to reproduce it. Which can be a problem because Face2Face works with just 5 minutes of video from the target. ”
No doubt for industry experts: these fake videos will improve over time. Especially since research is also progressing on the analysis of the voice. “Three or four years ago, it took almost two months of recording to create a synthetic voice that was quite similar. Today it is feasible in a few days of recording ”, confides William Simonin, CEO of Vivoka, a start-up specializing in voice recognition. Useful progress in many areas other than video faking (they help improve voice assistants in particular). But they could give deepfakes, by the way, an additional touch of realism.
What problems are these Deepfakes likely to pose? Are there tools to detect them? We will look at these topics in the next articles in this series dedicated to Deepfake.
The best software for deepfake
Despite Facebook’s official announcement regarding deepfake video, many users still want to know how to make this kind of video. A deepfake video is digitally edited video that allows audio and video files to be overlaid on top of other existing video. Most of the time, this video technique is used to make anyone say anything and change anyone’s face to another. Despite the prohibitions of video hosting sites on this type of content, there are some users who are still looking to know how to create a deepfake video. If you are one of these users, we offer you the following two tools to create a deepfake video.
Attention ! You will be responsible for your video so pay close attention to its content.
Create your own deepfake video
Deepfakes web (online)
Deepfakes web is a paid online service that allows you to create a deepfake video. However, if you are the kind of user who is always in a rush, be aware that with this tool it can take up to 4 hours to create a video. It supports most video formats: MP4, WEBM and MOV. In addition, the site imposes a limitation on the video length to 150 seconds and on the size to 50MB. To make your video successful, you will also be asked to provide a video in which the characters will be clearly visible. Now find out how to create a deepfake video with this online tool.
- Head over to the site and create an account.
- Then click on “Create a video”.
- Then upload a first video. This will be the basis of your video.
- Then, you will be asked to provide another video in which the same person from the first video appears but more clearly.
- Next, import the video which features the person you want to overlay on the first video.
- Finally, click on “Start Deepfakes! And wait for your video to finish processing.
Face Swap (sur mobile)
Face Swap is a mobile application that allows you to swap the faces of two people. Granted, it’s not quite a deepfake app, but it pretty much delivers similar results. However, with this app, you cannot import videos or photos from your mobile. You will be prompted to record or capture the video in real time. Besides, you can take photos, record videos, insert stickers and directly share your video or photo to social networks. Also note that the video requires the use of the Pro version of the application. Just follow the following steps to learn how to make deepfake video with this app.
- Go to Google Play Store and search for Face Swap app – Photo Face Swap. Install it on your mobile.
- After, open the application and select the option “Face Swap – Video” to record a video.
- Then place the faces of the two people in each frame and record the video. Simply press the “Stop” icon to stop recording.
- Finally, you can either directly share the video or save the video to your mobile.
Now you know how to create a deepfake video by following the steps provided in this article. Create your video at your own risk! You can also use BeeCut to make your changes!