Video @Scale 2024 - Day 1
summary
Panel and talks name concrete wins and engineering tradeoffs: generative video systems report real usage and speedups (Dream Machine: 25M users, 19.2s per 10s clip), research-grade models scale to MovieGen: 30B params using flow matching and multi‑resolution training, product teams ship creator features that make video at scale (AI Creator: 3M videos/month), and platform work covers translation, lip‑sync, provenance metadata, microservice encoding, and mobile upsampling tradeoffs. Pick this if you want numbers, latency fixes, and the actual pipelines used.
AI-written summary · about
chapters (11)
- 0:00Welcome to Video @Scale 2024
- 34:24Opening remarks
- 37:48Generative AI for video creation
- 1:02:01Safety in generative video models
- 1:14:34Movie Gen talk
- 1:37:38Captions' AI video generation
- 2:02:28Meta's AI translations pipeline
- 2:22:28Netflix's video pipeline
- 2:41:05Upsampling for video quality
- 3:00:00Q&A
- 3:27:59Closing remarks
related talks
44mAlmost No Magic, or How to Smoothly Deliver a Terabit Video Stream
49mBuilding Next-Generation Applications with EDA
08mStepan Pushkarev Interview
08mMicroservices at Scale
05mVladimir Bacvanski Interview
42mMicroservice Pipeline Architecture
02mThe Highlights
30mOn scaling Machine Learning microservices