Nude AI falls under the category of deep-learning models, because it relies on more sophisticated machine learning algorithms; in this case generative adversarial networks (GANs) to produce images. And these systems have been trained on large-scale datasets that can sometimes comprise of millions of images to learn patterns and structures in human anatomy. This AI learns to generate/reconstruct well-looking human forms, which results in NSFW content by looking at these images. According to TechCrunch, the rapid advancement of GANs means AI models can spit out images that look more realistic in just seconds running at 85–90% accuracy.
At its heart, this process relies on two neural networks working in concert. The first of these, the generator model is responsible for generating I-MJPEG images based on pre-Step 5 trained data. The second, the discriminator then checks these against real images to validate them proper appeal. The generator trains to become better at creating fake images, and discriminator aspires to get better with separating reals from fakes. Eventually, this leads to what can look like very realistic nude images that are sometimes impossible to distinguish as being fake photographs.
Technically, GANs are pixel-wise discriminative grown agents (sec 4.2) that generate real features of humans as well. This architecture models characteristics like skin tone, muscle and bone structure, body type / proportion (over pixel by pixel), allowing the generated images to retains these information. For example, Forbes said that some framework can generate 4k images of goods for an incredible level of pragmatic detail. The task takes usually just a couple of seconds, or at most a minute depending on the power of your system.
The training data for these models are commonly sourced from public datasets or private repositories. But, scraping these datasets from the internet raises some ethical questions. The New York Times profiled a case in which AI models were trained on social media images without the users' consent, raising conversations regarding breaches of privacy and abuse of data.
For my part, what impresses me most about nude AI is its potential to modify current images. That is, based on learned data regarding what human bodies look like under clothing, these AI systems can be used to take an image of a clothed individual and construct their naked version by infilling in the missing parts. Such technology can easily be used by abusive trolls to automatically generate images without consent, raising massive ethical and legal questions. The Verge reported that 80% of the deepfake content online is actually porn, you know because people are terrible and use AI tools wrong all the time.
Musk has previously spoken out about AI, saying : "AI doesn't have to be evil [...] If AI has a goal and humanity just happens in the way then […] it will destroy humanity as a matter of course without even thinking about it." The statement reflects the dangers posed by rampant AI misuse, particularly in controversial fields such as those with nude photos.
In generating AI-generated nudes, a second factor that will make this easier is the increasing computational power needed. Advanced AI models that can generate realistic, high-quality images are very compute-heavy and typically need dedicated processing power — such as GPUs (graphics processing units) or cloud infrastructure. Although generating these images is not free, the burgeoning consumer-grade AI tools have drastically reduced their prices: certain commercial platforms charges as little $10 per image.
Read more on: naked ai how does nude AI work and should we allow it. For a deeper dive into both the technology behind nude ai as well as considerations about privacy, ethics soceity visit Read More →