Long-to-Short Highlights

Turn one long piece of content (podcast / roundtable / talk show / interview / livestream replay) into multiple distinct-topic highlight clips in one shot, at a human rough-cut level:

Semantic selection: understands the content and extracts the topics worth their own clip — how many clips and how long each is are decided entirely by the content, never forced into fixed durations;
Jump cut (on by default): a second-pass compression inside each topic that keeps only the "spine" of substance, dropping repetition, back-and-forth confirmation, digressions, filler and stutters — squeezing rambling speech into a tight, information-dense clip;
Word-level precise cuts: every cut lands on a word boundary and snaps to the video frame, driven by word-level timestamps.

Positioned as a rough cut: it outputs editing decisions and project files you import into an editor to fine-tune; it can also render a rough cut directly.

Mix and match outputs (outputs): project files (each clip's .xml for Premiere by default; in project_formats you can also pick JianYing / CapCut / FCPX / OTIO / Tonghe .gtrk), selection report, and rendered rough cut (mp4).

Among these, .gtrk is the Tonghe unified editing contract (gtrk v1): add gtrk to project_formats and each clip produces it as one of the project files in that clip's files[] (alongside xml / JianYing / etc.). It can be imported into the Tonghe client for further refinement, or fed back as-is to Project File Export to convert to other NLE projects, or to Timeline Render for cloud rendering — the same edit structure throughout, with no re-parsing.

Create Task

Basics

Item	Value
Method	`POST`
Path	`/task/video_long2short`
Content-Type	`application/json`
Auth	`Authorization` header (raw API Key)
Billing	By input video duration (minutes)

Request Body

Field	Type	Required	Default	Description
`file_id`	`string`	Yes	—	Long video file ID (obtain via File Upload)
`language`	`string`	Yes	—	Content language (see supported list); a value outside the list returns an unsupported-language error
`output_language`	`string`	No	follows `language`	Output language for metadata (title / summary / score reason); same enum as `language`
`stt`	`object`	No	—	Pre-transcribed result (must include both word-level `word_tc_list` and sentence-level `sentence_tc_list`, see the stt object table below); skips internal ASR. Human scripts are not accepted
`main_topic`	`string`	No	—	Main topic, used as a soft hint for selection
`output_size`	`string`	No	`9:16`	Output canvas: `9:16` / `16:9` / `1:1`, or custom like `1080x1920`. Input may be landscape or portrait and is auto-fitted to the target canvas
`jump_cut`	`boolean`	No	`true`	Jump cut (intra-topic semantic compression). Off = each clip is a continuous rough cut with no redundancy removal
`duration`	`object`	No	`{}`	Duration control (see below). Purely semantic, no forced duration — only a safety cap and a soft hint
`split_screen`	`object`	No	`{enable:false}`	Optional vertical split-screen (see below): only multi-person on-screen segments are composed into multi-panel splits
`outputs`	`array`	No	`["project","report"]`	Output selection, any combination of `report` / `project` / `video`. `project` = produce project files per clip per `project_formats` (only Premiere's `.xml` by default); `report` = selection report; `video` = render per clip. Which project formats are produced (including `gtrk`) is decided by `project_formats`, not specified in `outputs`
`source_path`	`string`	⚠️ Conditional	—	The long video's local full path in your editing environment (e.g. `D:\clips\video.mp4`), written verbatim into project-file material references. Required when `outputs` includes `project` (the default outputs already includes `project`, so usually required); length ≤ 1024
`project_formats`	`array`	No	`["xml"]`	Effective when `outputs` includes `project`; specifies which project formats to produce (multiple allowed): `xml` (Premiere Pro) / `fcpxml` (Final Cut Pro) / `otio` (OpenTimelineIO) / `jianying` (JianYing) / `capcut` (CapCut) / `gtrk` (Tonghe `.gtrk`, openable directly in the client). Only `xml` is produced by default; list `gtrk` or other formats explicitly to get them
`struct_meta`	`object`	No	`{}`	Same as Project File Export: contains `nle_draft_dir` (your local draft folder; the legacy name `capcut_draft_path` is still accepted as an alias). Without it, JianYing / CapCut draft meta files (`*_meta`) cannot be produced

duration object

Field	Type	Default	Description
`pref`	`string`	`auto`	Duration preference hint: `auto` / `short` / `medium` / `long`; biases selection only, never forces re-cutting
`max_clip_sec`	`number`	`600`	Per-clip safety cap (seconds, 5 ~ 1800); prevents over-long clips only, does not limit clip count

split_screen object

Field	Type	Default	Description
`enable`	`boolean`	`false`	Enable vertical split-screen
`prefer_mode`	`string`	`auto`	Layout preference: `auto` / `single` (centered) / `multi` (multi-panel)
`orientation`	`string`	`auto`	Panel orientation: `auto` (randomized top-bottom/left-right across clips for variety) / `lr` (force left-right) / `tb` (force top-bottom)

stt object

⚠️ word_tc_list (word-level) and sentence_tc_list (sentence-level) are both required — neither may be omitted: the algorithm groups words into sentences by sentence_tc_list, so word-level only → 0 sentences → empty selection (silently returns nothing). All timestamps are in seconds (float).

Field	Type	Required	Description
`word_tc_list`	`array`	Yes	Word-level timestamps; element `{text: char, start_time: seconds, end_time: seconds}`
`sentence_tc_list`	`array`	Yes	Sentence-level timestamps; element `{text: sentence, start_time: seconds, end_time: seconds, speaker: speaker (optional)}`

{
  "stt": {
    "word_tc_list": [
      {"text": "投", "start_time": 12.30, "end_time": 12.46},
      {"text": "资", "start_time": 12.46, "end_time": 12.62}
    ],
    "sentence_tc_list": [
      {"text": "投资人卧底进喜茶，蹲了三个月。", "start_time": 12.30, "end_time": 15.80, "speaker": "嘉宾A"},
      {"text": "结果发现真正赚钱的根本不是奶茶。", "start_time": 15.80, "end_time": 18.42, "speaker": "嘉宾A"}
    ]
  }
}

Supported languages: Chinese zh-CN, Simplified zh-CHS, Traditional zh-CHT, English en-US, Japanese ja-JP, Korean ko-KR, French fr-FR, Spanish es-ES, Portuguese pt-PT, Russian ru-RU, Vietnamese vi-VN.

Request Example

curl -X POST https://api.ai-mcn.tv:10000/task/video_long2short \
  -H "Authorization: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "file_id": "81005495682916358",
    "language": "zh-CN",
    "output_size": "9:16",
    "jump_cut": true,
    "outputs": ["project", "report"],
    "source_path": "D:\\clips\\video.mp4",
    "project_formats": ["xml", "jianying", "gtrk"],
    "struct_meta": {"nle_draft_dir": "C:\\Users\\Me\\AppData\\Local\\JianyingPro\\User Data\\Projects\\com.lveditor.draft\\20260611"},
    "duration": {"pref": "auto"}
  }'

Success Response

{
  "code": 200,
  "msg": "success",
  "data": {
    "task_id": "537489015178400",
    "task_type": "video_long2short",
    "status": "queued"
  }
}

Query Task Result

Basics

Item	Value
Method	`GET`
Path	`/task/video_long2short/{task_id}`
Auth	`Authorization` header (raw API Key)

Response (output_result)

Field	Type	Description
`clips`	`array`	Highlight clips; each has publishing metadata + that clip's output files `files[]` (see below)
`report`	`object`	Selection report (unit count / clip count / jump_cut / split-material manifest, etc.)
`errors`	`object`	Local failure reasons for optional outputs (non-empty on partial success)

clips[] element

Field	Type	Description
`title` / `summary` / `tags`	`string` / `string` / `array`	Title, summary, tags (ready to publish)
`score` / `score_reason`	`number` / `string`	Spread-potential score (0~10) and reason
`highlight_words`	`array`	Highlight words with word-level timing `{text, begin_time, end_time}`
`genres` / `moods` / `themes`	`array`	Soundtrack genre / mood / theme keywords
`segment_count` / `total_duration_ms`	`integer` / `integer`	Sub-segment count after jump cut, and total clip duration (ms)
`text`	`string`	Retained text of the clip
`files`	`array`	The clip's output files. When `outputs` includes `project`, project files are produced per `project_formats` (only `["xml"]` by default), each with `type=project` and `filename=clip{idx}_*` (e.g. `clip0_premiere.xml` / `clip0_project.gtrk`; multi-segment concatenation after jump cut, target canvas = `output_size`); when it includes `video`, the render (`type=video`). Elements have `type` / `format` / `file_id` / `download_url` / `filename`

report structure

Field	Type	Description
`unit_count`	`integer`	Sentence units after word-stream re-segmentation
`selected_count` / `clip_count`	`integer`	Topics selected / clips actually produced
`jump_cut`	`boolean`	Whether jump cut was enabled
`jumpcut_used`	`integer`	Number of clips actually compressed
`degraded`	`boolean`	`true` if semantic analysis was temporarily unavailable and degraded (task still completes)
`split_manifest`	`array`	(when `split_screen=on`) split-material manifest: `{file_id, expected local path, covered clip}`

Success Response Example

{
  "code": 200,
  "msg": "success",
  "data": {
    "task_id": "537489015178400",
    "status": "completed",
    "progress": 100,
    "output_result": {
      "clips": [
        {
          "title": "Investor went undercover at Heytea for 3 months — the real money isn't the milk tea",
          "summary": "A frontline-research take on Heytea's real profit structure.",
          "tags": ["Business", "Tea drinks", "Research"],
          "score": 8.6,
          "score_reason": "Suspenseful open + counterintuitive conclusion; high completion and share potential",
          "highlight_words": [{"text": "根本不是奶茶", "begin_time": 15.8, "end_time": 18.42}],
          "segment_count": 6,
          "total_duration_ms": 61800,
          "text": "投资人卧底进喜茶，蹲了三个月……",
          "files": [
            {"type": "project", "format": "xml", "file_id": "537489015178411", "download_url": "/download/c0/537489015178411.xml", "filename": "clip0_premiere.xml"},
            {"type": "project", "format": "jianying_draft", "file_id": "537489015178412", "download_url": "/download/c1/537489015178412.json", "filename": "clip0_draft_content.json"},
            {"type": "project", "format": "gtrk", "file_id": "537489015178413", "download_url": "/download/c2/537489015178413.gtrk", "filename": "clip0_project.gtrk"}
          ]
        }
      ],
      "report": {
        "unit_count": 412,
        "selected_count": 9,
        "clip_count": 9,
        "jump_cut": true,
        "jumpcut_used": 7,
        "degraded": false
      },
      "errors": {}
    },
    "create_time": "2026-06-11T08:00:00Z",
    "update_time": "2026-06-11T08:06:42Z"
  }
}

Error Codes

Code	HTTP	Description	Resolution
`6013`	400	`file_id` / `language` missing	Add required params
`6004`	404	File not found	Check `file_id`
`6014`	400	File type mismatch (video required)	Upload a video file
`6016`	400	Invalid business params (`outputs` / `output_size` / `duration` out of range, `project` without `source_path`, `stt` missing word-level `word_tc_list` or sentence-level `sentence_tc_list`, etc.)	Fix per response detail
`6017`	400	Unsupported language	Use a value from the supported list
`6502`	401	Auth failed	Check `Authorization` header
`6201`	402	Quota insufficient	Buy a quota package or top up
`6202`	402	Balance insufficient	Top up in the dashboard

Limitations

Rough-cut positioning: outputs editing decisions / project files / rough renders, meant to be fine-tuned in an editor; jump cut is a hard-cut for density, and its brisk seams match short-video clip aesthetics.
Semantic, not duration-locked: clip count and length are decided by content; duration provides only a soft hint and a per-clip safety cap, with no forced re-cutting.
Word-level precision: internal ASR provides word-level timestamps, cuts land on word boundaries and snap to frames; a provided stt must include both word-level word_tc_list and sentence-level sentence_tc_list, otherwise sentences cannot be formed and the selection is empty.
Billed by input duration: by input video minutes, independent of clip count.
Core output guarantee: the task fails (auto-refund) only if the selection (clips) + report cannot be produced; project files and renders are derivative — single failures go to errors without failing the task.
Project material references: in the single-track project file, non-split segments reference the source_path original; ensure that path is valid in your editing environment. To get fully openable JianYing / CapCut drafts (with meta files), also provide struct_meta.nle_draft_dir. project_formats only produces Premiere's .xml by default; list other formats (including .gtrk) explicitly.
Split-screen (split_screen.enable=true): only multi-person on-screen segments are composed into vertical multi-panel splits and pre-rendered as split materials; outputs open up progressively as the capability matures.