How does Stable Diffusion turn noise into a picture?

7 Просмотры

  • Спасибо! Поделитесь с друзьями!

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

aii
Издатель
Stable Diffusion doesn't draw pictures — it un-blurs noise, fifty times in a row. Here's how that even works.

The training trick: take a real photo, add a tiny bit of static, then more, then more, until it's pure TV snow. Now train a small network to do one job only — given any noisy image, predict the noise that was added.

The flip: generation runs backward. Start with nothing but pure noise. Ask the same model where the noise is. It guesses, you subtract that guess, and the image is now slightly less random. Repeat about fifty times. Each pass peels off a layer of fog and a picture slowly emerges from the haze.

The steering wheel: you feed the model a text embedding — your prompt as numbers, like 'astronaut riding a horse'. At every denoising step the model peeks at those numbers and nudges the noise removal toward THAT idea. Text steers, the static gets sculpted.

Diffusion equals teaching a model to undo noise. Trained backward — generated forward.

Music: Markvard - Time [NCS Release] (NoCopyrightSounds)
https://ncs.io

#ai #diffusion #stablediffusion #generativeai #shorts #machinelearning
Категория
AI Картинки
Выполните вход или зарегистрируйтесь для отправки комментария.
Комментариев нет.