Video Material Search (for Smart Editing)

Search the copyrighted video material library with a single natural-language scene description or keywords, and get back clips ready to drop onto a timeline — including the best-matching in-clip time ranges (segments), a direct media URL, duration, orientation and fps. Returns synchronously, no polling required.

Designed for smart editing / auto-assembly: query material shot-by-shot from a storyboard, then stitch onto the timeline.

Search Endpoint

Basic Info

Item	Value
Method	`POST`
Path	`/task/video_clip_search`
Content-Type	`application/json`
Auth	`Authorization` header (pass the API Key directly)
Response	Synchronous search results (not an async task)

Request Body

Field	Type	Required	Default	Description
`query`	`string`	Yes	—	Scene description (natural language) or keywords
`top_k`	`integer`	No	`10`	Number of results, max `50`
`diversity`	`boolean`	No	`false`	De-duplicate near-identical shots
`request_id`	`string`	No	—	uuid, echoed back, for idempotency / tracing
`filters`	`object`	No	—	Editing-oriented filters, see below

`filters` Fields

Field	Type	Description
`min_duration`	`number`	Minimum clip duration (seconds)
`max_duration`	`number`	Maximum clip duration (seconds)
`orientation`	`string`	`landscape` / `portrait` / `square`, to match the output canvas
`min_width`	`integer`	Minimum resolution width (pixels)
`exclude_ids`	`array<integer>`	Material ids to exclude (already used in the same cut)

Request Example

curl -X POST https://api.ai-mcn.tv:10000/task/video_clip_search \
  -H "Authorization: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "silhouette of a person with a camera on a coastal rock at sunset",
    "top_k": 10,
    "diversity": true,
    "request_id": "550e8400-e29b-41d4-a716-446655440000",
    "filters": {
      "min_duration": 3,
      "orientation": "landscape",
      "min_width": 1920
    }
  }'

Success Response Example

{
  "code": 200,
  "msg": "success",
  "data": {
    "request_id": "550e8400-e29b-41d4-a716-446655440000",
    "recalled": 84,
    "results": [
      {
        "clip_id": 2693564728807342,
        "score": 0.87,
        "url": "https://cdn.example.com/nmcn/material/video_clip/raw/2693564728807342.mp4?expires=1750086000&sign=Xa1b2C3d4E5f6G7h8I9j0k",
        "cover_url": "https://cdn.example.com/nmcn/material/video_clip/keyframe/2693564728807342/cover.jpg",
        "preview_url": "https://cdn.example.com/nmcn/material/video_clip/preview/2693564728807342.mp4",
        "duration": 12.4,
        "width": 3840,
        "height": 2160,
        "fps": 25,
        "orientation": "landscape",
        "segments": [
          { "start": 2.0, "end": 8.5, "best": 5.2, "score": 0.91 }
        ],
        "matched": { "facets": { "time_of_day": ["dusk"], "scene": ["coast"] } },
        "note": "Dusk by the sea; a person on the rocky shore filming the water."
      }
    ]
  }
}

Response Fields

Field	Type	Description
`recalled`	`integer`	Total recalled (before filtering), for observability
`results[].clip_id`	`integer`	Material id. Note: this is a large integer that may exceed JS `Number.MAX_SAFE_INTEGER` (2^53) — JavaScript consumers should treat it as a string (e.g. quote the field at the response-text level before parsing), otherwise precision loss can collide ids of different materials
`results[].score`	`number`	Relevance score (after rerank, sorted descending)
`results[].url`	`string`	Temporary media URL (mp4, expires in 24h via `?expires=&sign=`; re-search for a fresh link after expiry)
`results[].cover_url`	`string`	Representative cover image (unsigned, never expires)
`results[].preview_url`	`string`	Low-res preview proxy (mp4, width ≤640, no audio track; unsigned, never expires). Ideal for list previews and a proxy-editing workflow: rough-cut with the small proxy first, then download the original via `url` when finalizing
`results[].duration`	`number`	Duration (seconds)
`results[].width` / `height`	`integer`	Resolution (pixels)
`results[].fps`	`number`	Frame rate
`results[].orientation`	`string`	Canvas orientation
`results[].segments`	`array`	In-clip matching ranges (sorted by `score`; first = best)
`results[].segments[].start` / `end`	`number`	Range bounds (seconds), ready for the timeline
`results[].segments[].best`	`number`	Best-matching timestamp within the range (seconds)
`results[].matched.facets`	`object`	Matched visual attributes (scene / time-of-day, etc.)
`results[].note`	`string`	Material text description

Error Codes

Code	HTTP	Description	Resolution
`400`	400	Malformed request body (body is not a valid JSON object)	Check the request body
`6502`	401	Authentication failed	Check the API Key in the `Authorization` header
`6401`	500	Upstream search service unavailable, or request parameters rejected by the search layer (e.g. empty `query`, invalid `orientation`/`filters` values)	Verify request parameters first; if correct, retry later
`6402`	500	Upstream search service timeout	Retry later

Determine success/failure from the code field in the response body, not the HTTP status code (top_k overflow does not error — it is clamped to 50).

Limits

top_k is capped at 50 (clamped automatically).
This endpoint takes a single query; for multiple shots, call it concurrently (batch input is a future capability).
Searches the copyrighted library only; material quality level is not differentiated.
The media url expires in 24h; re-search for a fresh link after expiry (cover_url / preview_url are unsigned and never expire).
segments are based on keyframes sampled every 1.5s, so localization accuracy is about ±1.5s.