Diffusion models
summary
StableDiffusion is a latent diffusion pipeline that maps a prompt through a frozen text encoder, repeatedly denoises a latent patch with a diffusion model, then decodes to a high‑res image; the common setting uses 50 steps. The open‑source model from StabilityAI supports image inpainting/outpainting, text‑guided variation and textual inversion to teach new concepts from a few photos. KerasCV provides a high‑performance implementation—enable KerasCV with XLA and mixed precision to run generation in seconds and the demo auto‑synthesizes an "American Pie" music video end‑to‑end.
AI-written summary · about
chapters (7)
- 0:00Intro
- 0:20Diffusion models
- 1:33Stable Diffusion
- 3:30Image Outpainting
- 4:08Stable Diffusion Architecture
- 6:36How to use it
- 8:41Demo video
related talks
01h 21mGenerative AI with Diffusion Models
10mLightning Talk: Diffusers: Bringing Cutting-Edge Diffusion Models to the Masses
41mBuilding Generative Image & Video models at Scale
49mGenerative Diffusion Models: Development, Training and Releasing Kandinsky 2.1
30mHow to train a large diffusion model
50mAI'll Be Back: Exploring Diffusion Models for Image, Audio & Video
46mHow to Make Stable Diffusion Generate Not Just Anime and Cats, but You
20mRealtime Video AI with Diffusion Models