FaceShifter technology
Let's take a quick look at how FaceShifter technology works?
FaceShifter is a two-step neural network framework for realistically replacing one person's face with another in a photo or video. This model supports partial concealment of the face behind various objects.
To date, FaceShifter is much better than previous approaches, preserving the traits of the target person in the generated image. Our technology uses two networks: one creates a fake face and takes the position of the head, facial expression, lighting, color, background and other attributes of the original photo for the fake.
The second network compares the photo created by the first network with the original photo. If it detects facial areas covered with hair, sunglasses or inscriptions, it eliminates possible errors in the first network.
Here we have framed photos of famous actresses in standard frames with other people, and you see how real everything turns out. Light and shadow lie just like on the original materials.
In the end, faces disappear behind the hair again, letters can be deciphered, and makeup sits where it is. And we get a result that's indistinguishable from the original.
Our neural network has been trained with hundreds of thousands of different photos and videos. We've reached a level where we can use even one photo. If you have several photos and they are of good quality, the result will be indistinguishable from the original. And in the end we get the perfect overlap of the face with the other person's body.
The idol of millions of Emmy Watson (Hermione).The photo of the face is analyzed by the neural network on 114 points and it allows to get a uniqueThe photo of the face is analyzed by the neural network at 114 points and it allows to get a unique deepfake
All you need is a picture of the face of the person we want to make the video on, one of the video scenes that are on the site and some time to calculate. At the end of the video rendering process, we have a unique porn video, which is very difficult to distinguish from the original and does not need to have a rich imagination to believe that the video is real.
We're not standing still, we're developing a sound substitution technology (SVT) that will replace the voice in the original video with a custom one. And soon your deepfake model will be able to speak with the voice you know.
We also test a system of visual patches that allow us to add tattoos, birthmarks and scars to the body of the model.
To date, FaceShifter is much better than previous approaches, preserving the traits of the target person in the generated image. Our technology uses two networks: one creates a fake face and takes the position of the head, facial expression, lighting, color, background and other attributes of the original photo for the fake.
The second network compares the photo created by the first network with the original photo. If it detects facial areas covered with hair, sunglasses or inscriptions, it eliminates possible errors in the first network.
Here we have framed photos of famous actresses in standard frames with other people, and you see how real everything turns out. Light and shadow lie just like on the original materials.
In the end, faces disappear behind the hair again, letters can be deciphered, and makeup sits where it is. And we get a result that's indistinguishable from the original.
Our neural network has been trained with hundreds of thousands of different photos and videos. We've reached a level where we can use even one photo. If you have several photos and they are of good quality, the result will be indistinguishable from the original. And in the end we get the perfect overlap of the face with the other person's body.
The idol of millions of Emmy Watson (Hermione).The photo of the face is analyzed by the neural network on 114 points and it allows to get a uniqueThe photo of the face is analyzed by the neural network at 114 points and it allows to get a unique deepfake
All you need is a picture of the face of the person we want to make the video on, one of the video scenes that are on the site and some time to calculate. At the end of the video rendering process, we have a unique porn video, which is very difficult to distinguish from the original and does not need to have a rich imagination to believe that the video is real.
We're not standing still, we're developing a sound substitution technology (SVT) that will replace the voice in the original video with a custom one. And soon your deepfake model will be able to speak with the voice you know.
We also test a system of visual patches that allow us to add tattoos, birthmarks and scars to the body of the model.