Anyone here with a bit of ML knowledge who can break it down?

[marnieswan]

So I’ve been digging around lately, trying to understand how Undress AI actually functions under the hood. Like, I get the concept — it uses AI to "predict" what someone might look like under clothes, but what exactly is going on technically? Is it GANs, neural networks, or some advanced inpainting method? I’m not interested in moral debates here — just the tech part. Anyone here with a bit of ML knowledge who can break it down?
 

[LarryCrow1992]

I’ve actually played around with similar ML models back when DeepNude made the rounds, and from what I can tell, Undress AI operates on a similar framework but with way more refined training and image synthesis layers. Most likely it uses GANs (Generative Adversarial Networks) fine-tuned on a very specific dataset, possibly with a conditional approach that allows it to map existing clothing contours onto predicted body structures. Some tools also apply deep inpainting, where missing regions (like underneath clothes) are "imagined" based on context from a massive dataset of real images. It’s not just swapping out textures — it’s rebuilding them pixel by pixel based on learned patterns. There’s some real complexity to it, especially in how lighting, skin tones, and anatomy are handled. You can see how it’s implemented more practically on their own site here: deepnude— they keep it pretty vague, but if you compare outputs to older tools, the progression is obvious.

[KarlLoder66]

I’ve wondered too how they get the skin tones to match lighting so well — that part seems more realistic now than earlier tools. Not saying it’s perfect, but clearly the tech has improved a lot.