← ClaudeAtlas

render-opslisted

Get a clip render out of ffmpeg and onto disk correctly — CFR at the exact fps, audio sliced by sample index (no drift across pieces), never bounding video with -t, frozen-frame detection, frame-count verification, two-pass loudnorm with encoder headroom measured on the ENCODED file at 48 kHz, and the namespace/delete rules for a workspace several clips share. Use before, during and after any render, and whenever a render freezes, drifts, clips its tail, or comes out too loud or too quiet.
vaibhav-barman/live-clipper · ★ 0 · AI & Automation · score 72
Install: claude install-skill vaibhav-barman/live-clipper
# RENDER OPS (ffmpeg edition) — the render, and getting it out intact > **THE ESSENCE:** a render that "succeeded" proves nothing. Frozen frames, drifting audio, a clipped > last word and a hot true peak all survive a successful exit code. **Measure the delivered file.** Adapted from the creator's Remotion render skill. **Kept:** frozen-frame detection, the audio-drift cause and fix, loudness measured on the encoded file, the frame-count check, the shared-workspace rules. **Dropped:** Remotion bundles, entry files and Node-API traps (the plugin renders with ffmpeg). ## Before you render 1. **One namespace per clip:** `work/<stream_id>/<clip_id>/` for every intermediate, `out/<clip_id>.mp4` for the deliverable. Never share a path between clips. 2. **Check who else is running:** `pgrep -fl ffmpeg`. Two renders writing one file produce a valid-looking mp4 with the wrong content (`moov atom not found` mid-session is the tell). 3. **Normalise the source first:** transcode the window to H.264, **CFR at the output fps (30)**, before any cutting. Never cut from a VFR stream recording. ## Audio drift across pieces Cutting N pieces out of an AAC track with `-ss/-t` snaps each to an AAC frame boundary (1024 samples), so audio lag **grows with every cut** (~0.6 frames per cut at 60 fps; +12 frames by 50 s originally) while the video stays frame-exact. **Invisible to the usual checks** — it gets misreported as a caption problem. **The fix:** 1. Decode the source audio to *