Video Analysis API Transcript and Visual Timeline

1. 1

   Send the link

   A YouTube, TikTok, Twitch, Kick or X link, or your own file. You never download or store the video.

2. 2

   Get the transcript

   Every spoken line with a start and an end time. Speaker labels and a time on every word are optional.

3. 3

   Get the visual timeline

   Events on screen, each with a start, an end and a description, plus scene boundaries.

4. 4

   Join on the timestamps

   Both results count seconds from the start of the video, so a quote lines up with the picture behind it.

curl -X POST 'https://api.deapi.ai/api/v2/audio/transcriptions' \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json' \
  -H 'Authorization: Bearer YOUR_API_TOKEN' \
  -d '{
    "source_url": "https://www.youtube.com/watch?v=VIDEO_ID",
    "model": "WhisperLargeV3Ct2",
    "include_ts": true
  }'

# → {"data": {"request_id": "c08a339c-…"}}
# Poll GET /api/v2/jobs/{request_id} or pass webhook_url.

curl -X POST 'https://api.deapi.ai/api/v2/videos/descriptions' \
  -H 'Accept: application/json' \
  -H 'Authorization: Bearer YOUR_API_TOKEN' \
  -F 'video_url=https://www.youtube.com/watch?v=VIDEO_ID' \
  -F 'model=Marlin_2B' \
  -F 'preset=balanced'

# → {"data": {"request_id": "7d1e52aa-…"}}
# Same link, same job endpoint, same webhook.

// 1 · transcript
{
  "language": "en",
  "segments": [
    { "start": 2712.4, "end": 2716.9,
      "text": "Quick shout-out to tonight's sponsor." }
  ]
}

// 2 · visual timeline
{
  "events": [
    { "start": 2712.0, "end": 2721.0,
      "description": "The streamer lifts a drink can
        toward the camera and smiles." }
  ],
  "scenes": [
    { "idx": 41, "start": 2690.0, "end": 2745.0,
      "text": "Streamer at a desk, gameplay paused." }
  ],
  "meta": { "duration": 7214, "preset": "balanced" }
}

Reference: transcription and video description.

• Fast

  The quick pass

  Small frames over long windows. The cheapest way to index long video.

  From $0.073 per hour of video

• Balanced · default

  The middle setting

  Used when you leave the preset out. A middle point between price and how much of each frame is read.

  From $0.104 per hour of video

• Detail

  The close read

  Large frames over short windows. Reads more of what is on screen, including small text and brand names.

  From $0.216 per hour of video

• Brand and sponsor tracking

  Check the mention against the picture

  Search the transcript for the sponsor's name. Read the same seconds on the visual timeline to confirm the product was on screen.

• Video search

  Search inside the video

  Index both results, and a query like “unboxing” or “whiteboard” returns a time range, even when nobody said the word.

• Highlights and clipping

  Cut on what happens

  Scene boundaries and event times are your cut points. Pick the events you care about and clip from start to end.

• Agents and RAG

  Give a model the whole video

  Hand a language model two timestamped text files in place of a 2-hour video. It can answer questions and cite the second it quotes.

Balanced is the default preset. A 2-hour Twitch VOD with a balanced timeline costs $0.75. Speaker labels or a time on every word add 50% to the transcript. Source metadata adds $0.005 per request. Longer TikTok clips cost more per clip, and the visual timeline takes TikTok clips up to 10 minutes. All rates on the pricing page.