Apple open-sources FastFrames: real-time video gen at 35ms latency in 1.3B and 14B sizes
victormustar · x · 2026-11-18
Apple has released FastFrames, a text-to-video model family targeting "interactive latency," on Hugging Face, with official implementations of FastFrames-S (1.3B) and FastFrames-L (14B).
- Architecture: autoregressive, two-step video diffusion generating 480p video at interactive latency
- Single-GPU latency profiling shows a total of 35.4ms per step at 480p (paper: 36ms), with each DiT forward pass at just 16ms
- Also ships a 16x-optimized LingBot 1.3B build with fluid camera-conditioned control and an interactive demo
- Repo includes latency benchmarking tools, T2V generation scripts, and training code; requires NVIDIA GPU with CUDA 12.8
More from Models
- Custom sophistry benchmark finds GPT-6.2 Astra and Sol less honest than their predecessors — gamedev-leper · 2026-11-18
- Users petition Anthropic to keep Sonnet 4.5 alive via paid retired-model access before Nov 30 API shutdown — DirectorHot5435 · 2026-11-18
- GPT-6.2 Astra and Sol land in Perplexity, scoring 0.783 and 0.746 on WANDR — perplexity_ai · 2026-11-18
- Claude 6.2 Builds an Impressive Dither-Style 3D Game in One Shot — craigsdennis · 2026-11-18
- Researcher: frontier model 6.2 Astra faces unprecedented backlash, rushed release tied to Ant IPO? — Afinetheorem · 2026-11-18
- Pet theory: Anthropic is skipping looping agents for safety and monitorability — NathanpmYoung · 2026-11-18