Scene logs for 20 reference ads: by hand vs $0.30 per analysis

Twenty reference ads through Sume's legacy video-analyses route would cost $6.00 at $0.30 per accepted analysis, and each returns typed scenes with summaries, shot type, on-screen text and stills. The docs also say not to start new work on this surface: it returns 410 video_analysis_retired on the development API and stays on in production only until a later cleanup (Video analyses).

For new clip inspection the docs point to video inspect, which probes and takes stills and can transcribe.

Cost at volume

What the by-hand log has to cover

Each row of a manual scene log maps to a field in a returned scene.

Limits to respect

max_scenes accepts 2 to 40 (default 24). Clips over about 90 seconds succeed with a low_confidence_long_video warning, and the hard cap is 300 seconds. YouTube URLs return 422 unsupported_source, and longer videos give less accurate results than short clips.

curl -X POST https://api.sume.com/v1/video-analyses \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: analysis-001" \
  -d '{"video_url":"https://media.sume.com/artifacts/example/ad.mp4","max_scenes":24}'

Where to go for new work

Use POST /v1/video-inspect for probe and stills, or reference ingest for shots, text and audio facts. Run the production create above only if you have an existing integration that depends on it.