Removes dependency on MotionEye or any NVR software. Everything is now
CLI/ffmpeg configured entirely from sky-cam.conf.
capture.sh
Long-running systemd service (one per camera) that pulls frames from the
camera's RTSP stream at CAPTURE_INTERVAL seconds, writing HH-MM-SS.jpg
into BASE_DIR/<cam>/YYYY-MM-DD/. Restarts at midnight for the new date
directory; auto-reconnects on camera disconnect.
sunrise-audio-capture.sh
One-shot service triggered at 03:00. Waits until (sunrise minus
SUNRISE_PRE_MIN minus AUDIO_PRE_BUFFER_MIN), then records the RTSP audio
stream for the full sunrise window. Output deleted after mixing.
daily_sunrise_video.sh
Step 3 now mixes in sunrise-audio.m4a when AUDIO_ENABLED=true and the
file exists. Audio is centred on actual sunrise time at natural speed
while the video plays as the sped-up timelapse. Uses filter_complex
when audio is present (can't combine -vf with -filter_complex).
install.sh
Generates sky-cam-capture-<cam>.service for each camera with CAM_RTSP_<cam>
set, and sky-cam-audio-capture.{service,timer} when AUDIO_ENABLED=true.
sky-cam.conf
New settings: CAM_RTSP_<cam>, CAPTURE_INTERVAL, AUDIO_ENABLED,
AUDIO_PRE_BUFFER_MIN, CAPTURE_AUDIO_BITRATE.
https://claude.ai/code/session_01C4jbd3waXG3eKZYbGUjLUQ
Sunrise:
- Overlay failure no longer exits 1; the speed-only video is renamed to the
expected final path so OnSuccess still fires and the upload still happens
- sunrise2mm.py now notifies on every upload outcome (success, post failure,
file-not-found) via notify.sh so the user always knows whether the video
reached Mattermost
Montage:
- Music+overlay failure no longer exits 1; the sped video is promoted to the
*-Montage.mp4 path so year-end-join can still include this movement (as a
silent, no-overlay clip) and the user is notified of the degraded output
- Year-end join trigger is now outside the success/failure branch so it fires
in both cases
https://claude.ai/code/session_01C4jbd3waXG3eKZYbGUjLUQ
Each pipeline now saves intermediate files permanently before the step most
likely to fail, so a partial failure leaves a recoverable artifact:
- daily_sunrise_video.sh: speed-adjusted video (*-sped.mp4) survives if the
drawtext overlay step fails; deleted automatically on success
- montage-mvt.sh: speed-adjusted video (*-Sped.mp4) survives if the
music-mux + attribution overlay step fails; deleted on success
- 4-seasons.sh / montage-mvt.sh / year-end-join.sh: explicit ffmpeg error
handling with targeted notify.sh calls naming the failed step and the
surviving artifact path
- montage-mvt.sh / 4-seasons.sh / year-end-join.sh: clip-duration-sum vs
music-duration drift check warns when cumulative float error exceeds 2 s
https://claude.ai/code/session_01C4jbd3waXG3eKZYbGUjLUQ
daily_sunrise_video.sh:
- Fixed: was using undefined \$MOVIES_DIR (now \$BASE_DIR/movies)
- Fixed: sunrise_overlay.py was called but never existed — would fail daily
- Fixed: speed-adjust step had no explicit codec (-c:v libx264 missing)
- New: sunrise time overlaid as stacked vertical characters on the right
side using ffmpeg drawtext — no Python/Pillow dependency
- New: upload removed from script; now triggered by systemd OnSuccess=
- Combines speed-adjust and overlay into one ffmpeg pass (was two)
- SUNRISE_OVERLAY_OPACITY from conf controls transparency
install.sh:
- sky-cam-sunrise.service gains OnSuccess=sky-cam-sunrise-upload.service
so upload only runs when video creation succeeds, with separate log
- New sky-cam-sunrise-upload.service runs sunrise2mm.py
sunrise2mm.py: add shebang + chmod +x so systemd ExecStart= calls it directly
sky-cam.conf: add SUNRISE_OVERLAY_OPACITY=0.45
https://claude.ai/code/session_01C4jbd3waXG3eKZYbGUjLUQ
Adds six CRF vars, AUDIO_BITRATE, and three MONTAGE_* overlay durations.
Values match previous hardcoded defaults so behavior is unchanged.
The same CRF appeared in several scripts but with intentionally different
values per purpose (28 for draft intermediates, 26 for polished output).
Each script's CRF is now its own variable in conf, so changing one never
silently changes another.
https://claude.ai/code/session_01C4jbd3waXG3eKZYbGUjLUQ
Claude