Diffusion Models Explained: DDPM, DDIM, Latent Diffusion and Stable Diffusion

5 Просмотры

  • Спасибо! Поделитесь с друзьями!

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

aii
Издатель
How does AI turn pure noise into a picture? This hand-drawn explainer walks through diffusion models: the forward process that drowns an image in noise over about 1,000 steps, the U-Net that learns to predict and remove that noise, the June 2020 DDPM paper by Jonathan Ho, Ajay Jain and Pieter Abbeel, DDIM's 10 to 50 times faster sampling (Song, Meng and Ermon, 2020), latent diffusion from Robin Rombach and colleagues in Munich (December 2021), and the August 2022 open-weights release of Stable Diffusion - plus how the CLIP text encoder and cross-attention let your prompt steer every denoising step.

Chapters:
0:00 From static to picture
0:24 The forward process
0:49 The reverse process and the U-Net
1:15 DDPM: the 2020 recipe
1:36 Slow sampling and DDIM
2:14 Latent diffusion
2:54 Stable Diffusion and open weights
3:13 Text encoders, CLIP and cross-attention
3:48 Why diffusion won

Part of AI Model Families Explained - one hand-drawn explainer per AI lab.
#DiffusionModels #StableDiffusion #AIExplained
Категория
AI Картинки
Выполните вход или зарегистрируйтесь для отправки комментария.
Комментариев нет.