Apple open-sources FastFrames: interactive-latency video gen at 35ms per frame
Jimantha · x · 2026-11-19
Apple AI ML research released FastFrames, arguing latency (input-to-RGB-output) matters more than FPS for interactive video: FastFrames-1.3B generates a frame in 35ms on one B200, FastFrames-14B in 45ms on 8x B200, and LingBot-1.3B is accelerated to 50ms for responsive camera control.
All code and model weights are open-sourced, with the paper and project page on Hugging Face — an open baseline for real-time interactive video / world-model applications.
Related event: Apple Open-Sources FastFrames Video Generation Models(3 posts)→
More from Multimodal
- Blue Moonlit Nights: A Dreamy Set of AI-Generated Nocturne Images — Kyrannio · 2026-11-19
- Peter Gostev benchmarks GPT-6.2 Astra vs Sol across 3D, games, slides and apps — arena · 2026-11-19
- Fully AI-generated fashion runway video wows Reddit with realistic catwalk finale — Ashayaki · 2026-11-19
- Animated AI ASMR Episode 2: Ollivander's Wand Shop — Nthorpe1996 · 2026-11-19
- FaceRefine Plus v0.2.0 for MiniMax H3: 2.7x faster, one-dropdown V1/V2 modes — AnniCodex · 2026-11-19
- Fans make an AI music video begging NVIDIA's board to 'Bring Jensen Back' — AIandDesign · 2026-11-19