Lightning Talk: Diffusers: Bringing Cutting-Edge Diffusion Models to the Masses
summary
Diffusers is a PyTorch library that loads Hugging Face checkpoints to run image, video, audio and 3D diffusion models with minimal code — examples generate an “astronaut in a jungle” image and a Darth Vader sequence guided by ControlNet SD-OpenPose. It supports Stable Diffusion XL and many auto-pipelines, exports videos frame-by-frame, and uses torch.compile() plus SDPA for large speedups (notably on low batch sizes). Community UIs and mobile toolkits are already built on it.
AI-written summary · about
chapters (7)
- 0:00Introduction to diffusers
- 0:39How diffusion models work
- 2:22Getting started with code
- 4:49Beyond text-to-image
- 5:55Advanced control with ControlNet
- 7:23Pipeline and model flexibility
- 8:44Performance and community
related talks
10mDiffusion models
25mtorch.compile and Diffusers: A Hands-On Guide to Peak Performance
19mYou Might Not Need 50 Diffusion Steps
41mBuilding Generative Image & Video models at Scale
01h 21mGenerative AI with Diffusion Models
50mAI'll Be Back: Exploring Diffusion Models for Image, Audio & Video
01h 16mOptimizing workloads with Transformers and Diffusers
12mEfficient Workflow Serving for Diffusion Models with Many Adapters