fastframes-s-live / README.md
linoyts's picture
linoyts HF Staff
Stream: render every edit (no debounce), 0.5 s segments, 0.6 s lead; reuse KV cache
4675f39 verified
|
Raw History Blame Contribute Delete
1.31 kB

A newer version of the Gradio SDK is available: 6.32.0

Upgrade
metadata
title: FastFrames-S Live
emoji: 🎞️
colorFrom: yellow
colorTo: gray
sdk: gradio
sdk_version: 6.30.0
python_version: '3.12'
app_file: app.py
pinned: false
license: other
license_name: apple-amlr
license_link: https://huggingface.co/apple/FastFrames-S/blob/main/LICENSE
startup_duration_timeout: 1h
short_description: Endless text-to-video that follows your prompt live
models:
  - apple/FastFrames-S
  - zhuhz22/Causal-Forcing
  - Wan-AI/Wan2.1-T2V-1.3B

Real-time, prompt-reactive text-to-video with apple/FastFrames-S. Press Go live and keep editing the prompt; the stream follows it.

  • One held ZeroGPU session per user streams 0.5 s segments; prompt edits reach it through a per-session file.
  • Every prompt edit regenerates the scene from frame 0 with the same seed, so the new prompt is followed exactly and the shot stays comparable.
  • An idle prompt keeps streaming past 5 s with a sink + rolling KV window (keys re-rotated so RoPE positions stay in 0..20) and gets a fresh scene every 20 s (adjustable) to stop long-rollout drift.
  • Inference code is the official ml-fastframes package (commit 6044a56), vendored unchanged in fastframes/. See LICENSE and ACKNOWLEDGEMENTS.