How Does AI Turn Static Into a Picture?

1 Просмотры

  • Спасибо! Поделитесь с друзьями!

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

aii
Издатель
Many AI image generators start every picture as pure static. In training, real pictures are buried in noise, from a little to total static, and the model learns to estimate the noise that was added. To make something new, it starts with fresh static, and your prompt steers each guess: remove some noise, guess again. Classic Stable Diffusion repeats this for dozens of steps (a typical sampler uses 20-50) on a compressed 64x64x4 latent, 48x fewer values than a 512x512 RGB image, then decodes it into pixels. Same prompt, different starting noise? You can get a whole new picture.

The particles are an illustration of changing numbers, not a real model run. Other generators (flow-matching and autoregressive models) work differently.

Sources:
DDPM (Ho et al., 2020): https://arxiv.org/abs/2006.11239
Latent diffusion (Rombach et al.): https://arxiv.org/abs/2112.10752
Stable Diffusion v1 model card: https://huggingface.co/CompVis/stable-diffusion-v1-4
Hugging Face Stable Diffusion explainer: https://huggingface.co/blog/stable_diffusion

#shorts #ai #stablediffusion #aiart #howitworks #tech #machinelearning
Категория
AI Картинки
Выполните вход или зарегистрируйтесь для отправки комментария.
Комментариев нет.