Reka and NVIDIA demo real-time steerable video gen: 720p 24fps, 11.8x faster on one H100
ArashVahdat · x · 2026-09-04
Reka AI and NVIDIA unveiled a real-time version of Reka's video generation model: 30B parameters, steerable mid-stream via natural language, 720p at 24fps on a single H100, 11.8x faster with no major quality drop in blind evals. It's built on NVIDIA's open-source FastGen framework, a PyTorch stack supporting ≥10B-scale training across T2I/I2V/V2V with distillation methods like consistency models and DMD. Closed beta is open; the team frames it as a step toward World Language Action Models.
Related event: Reka and NVIDIA Unveil Real-Time Steerable Video Generation(2 posts)→
More from Infra
- DeepMind infra lead calls Gartner's 'compute surplus' claim nonsense: we're massively constrained — appenz · 2026-10-21
- FastVideo open-sources MiniMax FastH3 Omni with training code, 12x faster on Blackwell — multimodalart · 2026-10-21
- Anthropic's draft IPO filing: $16.3B H1 revenue, $80B net loss, $750B compute plans — kimmonismus · 2026-10-21
- dphnAI's distributed inference network served 32B DeepSeek tokens in 24h, ~6% of Venice's daily volume — Ronangmi · 2026-10-21
- Jensen Huang: every company will have two factories, one real and one digital — round · 2026-10-21
- DeepSeek V4 called 'god-tier infra, 4th-tier ML': 7 tricks stacked, 20 patches to fix them — teortaxesTex · 2026-10-21