AI Doesn't Copy Pictures. Here's What It Really Does

1 Просмотры

  • Спасибо! Поделитесь с друзьями!

    Вам не понравилось видео. Спасибо за то что поделились своим мнением!

aii
Издатель
Most people think AI image generators stitch together pieces of pictures from the internet. They don't. There's no library of images inside.

A model like Stable Diffusion was trained on about 2 billion images, yet its weights fit in about 2 gigabytes. That's roughly one byte per image, far too little to store pictures. What it keeps are patterns.

Here's how it actually draws: it starts from pure random static. Step by step, it removes noise, guided by your prompt, until a picture appears. This is called diffusion. It learned the skill by taking billions of images, burying them in noise, and practicing digging them back out.

That's also why the same prompt gives a different image every time: new static, new picture.

(Research has shown models can occasionally memorize images that appear many times in their training data, but there's no stored library to copy from.)

#ai #aiart #science #technology #machinelearning #diffusion
Категория
AI Картинки
Выполните вход или зарегистрируйтесь для отправки комментария.
Комментариев нет.