294 likesNew · 3d old
Apple releases FastFrames-L: 14B text-to-video model at interactive speed
What is FastFrames-L?
FastFrames-L is a 14-billion-parameter autoregressive diffusion model that turns text prompts into 480p video in roughly 35 ms per frame, delivering near-real-time generation. The repo includes latency profiling tools, a LingBot demo, and evaluation scripts.
Who it is for
Developers and creators who need fast text-to-video generation can experiment with it today.
How to try it
CUDA_VISIBLE_DEVICES=0 uv run --extra lingbot --extra demo python apps/lingbot/demo.py --gpus 1
The story so far
Early Signal first spotted apple/FastFrames-L on November 19, 2026, with 294 likes. On its busiest day it gained 98 likes. The project is 3 days old. Early Signal files it under Video, Image, Inference.
Questions about FastFrames-L
- How do I install FastFrames-L?
- Use one of these, depending on your setup: uv sync · uv run python evaluation/latency/benchmark.py ff-l --gpus 8 · CUDA_VISIBLE_DEVICES=0 uv run --extra lingbot --extra demo python apps/lingbot/demo.py --gpus 1
- What do I need to run FastFrames-L?
- Linux x86-64 with an NVIDIA GPU and a CUDA 12.8-compatible driver