AI-Generated Nudes are based on machine learning algorithms, mainly generative adversarial networks (GANs). A GAN is formed by a generator and a discriminator. The generator…It generates the images with respect to data input (Z) and discriminator tries to evaluate the given images on whether they are real or not. In this way, the generator keeps getting better at its output and eventually produces real looking images. GANs are normally empowered with large datasets, ranging from thousands to (even) millions of images. The datasets are essential to teach the AI how to create lifelike figures of humans — with skin texture, body-proportions and all kinds of lighting effects in tow.
The training of such models can take anywhere from days to even weeks depending on the model complexity and available computational power. A standard GAN model alone may require as much as 500 hours of GPU processing power to achieve a high level of realism. The generated images can have different resolutions and quality depending on the size of the training data set and computational resources used. Better dimensioned images demand more processing power and cost an additional amount of data, what makes the whole thing even more time-consuming.
While some AI-rendered nudes are modified from existing photos, others are made entirely from scratch. The AI can even morph various aspects like body types, skin color, and face shapes to generate new images that don't correspond to any actual person. This practice also poses ethical concerns, as it blurs the line between art, privacy and consent. The creation of non-consensual explicit material was a similarly hot button issue in 2019, and the rise of deepfake technology sparked fears around the misuse of AI to manipulate such content illegally; this prompted increased legal action taken against creators and a strengthening of regulation across much of the world.
In one infamous case this year, a deepfake app was released that let users upload non-nude photos of women and have it spit back out what they looked like naked. The app was quickly removed, but it raised awareness among the public that AI can cause great harm if used irresponsibly in this area. This is an example of just how massive the deepfake issue has persisted – one cybersecurity firm reported that in 2021, 96% of all pornographic video detected online were deepfakes.
Regarding image quality, ai generated nudes are able to reproduce a lot of detail in terms of skin texture, lighting and anatomy. Certain AI models can produce 1024x1024 pixels images, with clear and actually looking output. These images are sometimes in need of some tweaking as the AI may produce unrealistic or even distorted results, particularly with complex areas such as hands or facial expressions. While the technology is getting better,remains along way offfullhuman features.
Elon Musk said, “With AI, we are summoning the demon”, he continues on to write more about other things we can do with it (“we have all of these dangerous scenarios”) but also reinforced it by acknowledging that when developing this technology that it “could be like summoning the demon”. It highlights the promise and pitfalls in creating realistic AI-style nudes. Not only is this capability dangerous due to the obvious ethical impications surrounding consent and misuse, but it will essentially create a near unlimited supply of fake people.
For more detail and include technical advancement / impact of this technology, please refer to ai deepnude which explains the power & real applications present in AI image generation.