DeepNude is an application that uses artificial intelligence and neural networks to remove women's clothing from images and recreate ultra-realistic nude. It is an evolution of the so-called 'deep fakes', a new algorithm that anticipates the great potential of this technology as well as its impact on society.

It works in a simple way. DeepNude allows you to upload an image of a woman in clothes and create a new image of the same person but where the shirt, bra, bikini or any garment that covers her parts has been removed. An image marked 'fake' by the program itself, but with a rather realistic result and could well pass through a real image.

An app that only recreates images of women

Accessing DeepNudes is very simple as it is available online through a website. There is a free version that allows you to transform the photos with some limitations and adds a visible watermark and then a payment version that allows you to export to full resolution, reduces the brand to a "fake" at the top and allows you to better modify the image. This payment version initially had a cost of $50 as Motherboard reports, but due to the attention generated it has raised the cost to $99.99. Being able to pay by card or by cryptocurrency.

The joy of learning something new is a blessing we should always cherish and celebrate. Brooklyn was very excited... Http: / / t.co

— Todd Gulluscio, CMAA Tue Nov 24 02:22:30 +0000 2020

The platform is available via the web but it is through an application for Windows and Linux that the user can upload images. The creator of the app says he is also working on an app for Mac. No mobile app is available at the moment.

That said, the app manages to create nudity in seconds. It's so simple it really scares. Fake nudity has existed on the internet for years, but with deepfakes we enter a new phase, much more dangerous as it leaves it to anyone to generate such an image. You Don't need editing skills anymore, it's the algorithm that's in charge. An algorithm that only works with women, since when we test it with an image of a man instead of showing a penis, it adds again a vulva and women's tits.

As the author shows in his official account, DeepNude is based on the X-ray vision glasses they promised to see through their clothes. The "superpower you always wanted to have", according to the official slogan of the application.

We are "facing an invasion of sexual privacy," says Vice Danielle Citron, a law professor at the University of Maryland who recently testified in Congress for the use of deepfakes. And the arrival of these fake nudity generated by algorithm represents a new chapter in the dangers of deepfakes.

DeepNude could be used for revenge porn even though those images never really existed without clothes. The result is so realistic that the impact it could have is similar. An application created "for fun and curiosity" as explained by the author but which generates a whole series of doubts about privacy and links machismo to technology again. Why does the algorithm only work with women? "because images of naked women are easier to find online," explains the author to Vice. To which he adds that he hopes in the future to create a version with men as well.

Based on an open source algorithm

DeepNude data are stored locally and not uploaded to the cloud indicates the website. The author of the application did not want to reveal his real identity, although he claims to be 'Alberto' and from Estonia. It does explain more details about how DeepNude's algorithm works. The software would be based on pix2pix, an open source algorithm developed by the University of California, Berkeley in 2017.

It is an algorithm that uses neural networks to deal with a large database of images. In this case, more than 10,000 images of naked women that the programmer would have used to "train" artificial intelligence.

While deep fakes videos take hours to render, these images are generated in less than a minute. "the networks are multiple, each one has a different task: to locate the clothes, select them, speculate with the positions of the body and render it", a process that the author explains "can be done in 30 seconds on a normal computer and can be improved in the future".