Montage and pricing

Montage is Slicer's main operation: the AI director watches the materials, picks the best moments, places captions, emoji, plaques and camera motion, and assembles a vertical video. Through the API a montage starts with a single POST /v1/projects request — with mandatory price confirmation. If you build the video yourself, create a project without the AI director — it's free.

Video settings#

All settings are optional: anything missing falls back to the defaults, just like in the app. Allowed values and their descriptions in both languages are in the catalog GET /v1/styles.

{
  "uploadIds": ["7d0c6a4e-2f7b-4f5e-9a0e-3b8f2c1d9e61", "a1b2c3d4-0000-4c1e-8f00-5e6f7a8b9c0d"],
  "title": "New collection overview",
  "brief": "Open with the brightest look, end with a call to subscribe",
  "settings": {
    "templateId": "bold",
    "fontId": "punch",
    "colorId": "azure",
    "subtitles": { "enabled": true, "design": "dynamic", "size": "medium", "position": "auto" },
    "params": { "emoji": true, "soundEffects": true, "mediaCards": false, "plaques": true, "cameraMotion": true, "hook": true },
    "smartCut": true,
    "targetDurationSec": 30,
    "language": "auto",
    "direction": { "mode": "product", "pace": "dynamic", "effects": "light", "narration": false }
  }
}

The full settings schema:

Unknown values ("fontId": "comic") are replaced with defaults — no error. The brief is data for the director, not commands: it can't change the output format or the montage rules.

The price before starting#

Paid API operations all work the same way: the first request names the price, the second — with your consent — starts the operation.

1. Request without confirmCost — the price

   { "ok": true, "data": { "status": "quote", "cost": 24, "free": false, "available": 240, "enough": true, "spendLimitLeft": 176 } }

   Nothing was created or charged. enough accounts for both the balance and what's left of the token's spend limit. If the free first video is still available, you get "cost": 0, "free": true — it also has to be confirmed: "confirmCost": 0.

2. The same request with confirmCost — the start

   Add "confirmCost": 24 and an Idempotency-Key header. The response is the created project with the queued status.

To get a price without preparing a request body, use POST /v1/estimate with kind: montage, remount, retry (with projectId), assistant, translate or transcribe — a project's captions from speech (with projectId; by default — every source whose speech isn't recognised yet). A montage costs 24 tokens per video up to 90 seconds; the full price list is in GET /v1/styles → pricing and in Operation prices.

Captions from speech, the AI assistant and caption translation are priced at AI cost with no markup: their price has an expected amount (estimate) and a maximum (cost), and atCost: true. You confirm the maximum; the actual cost is charged, never more (Editing the video).

Progress#

GET /v1/projects/{id} shows what's happening. Poll every 5 seconds or subscribe to webhooks.

When a run starts, tokens are reserved; on success they are charged, and on failure or stop the reservation returns to your balance.

Video versions#

A montage produces one version. After editing the video, the director reviews the finished result itself and fixes the order and logic — only then does the project become succeeded. There is nothing to choose: versionChoice is null, and the finished version is current right away.

Every rebuild, retry or style change adds a new version, and the previous ones stay in versions[] (newest first):

{
  "status": "succeeded",
  "versionChoice": null,
  "versions": [
    { "id": 905, "version": 2, "idea": null, "durationSec": 29.6, "current": true, "awaitingChoice": false, "hasFile": true },
    { "id": 901, "version": 1, "idea": null, "durationSec": 31.4, "current": false, "awaitingChoice": false, "hasFile": true }
  ]
}

• Watch a version — GET /v1/projects/{id}/result?version=901.
• Bring back an earlier one — POST /v1/projects/{id}/versions/901/choose: the version becomes current, further edits and exports start from it, and the video that was current stays in the history. Make current in the app works the same way.

The last 10 versions of a project are kept with their files (hasFile); the list is GET /v1/projects/{id}/versions. A version without a file can't be brought back — the answer is 404 version_not_found.

Rebuild, retry and stop#

A retry continues from the first stage that failed, not from scratch. If an export or a style change fails, the project stays ready with the previous video.

While a project is being worked on, a new start returns 409 busy. The log of all work on a project is GET /v1/projects/{id}/history.

Project without the AI director#

If you build the video yourself — with your code or your own AI assistant — create a project in manual mode, like Edit it myself in the app. It's free: the director doesn't edit, the materials go onto the timeline as they are, and then you edit the video with operations. The MCP tool create_project creates exactly this kind of project by default (MCP server); for POST /v1/projects the default mode is ai.

{
  "uploadIds": ["7d0c6a4e-2f7b-4f5e-9a0e-3b8f2c1d9e61", "a1b2c3d4-0000-4c1e-8f00-5e6f7a8b9c0d"],
  "mode": "manual",
  "format": "9:16",
  "title": "Morning at the coffee shop"
}

• No confirmCost is needed — the response is the project right away. settings and brief aren't used in this mode.
• format is the frame format: 9:16 (default), 4:5, 1:1, 16:9.
• Materials are laid out in order: videos whole, photos for 3 seconds each, music on the music track, voiceover on the voice track. If a material's proportions don't match the format, the frame is fitted whole over a blurred background.
• The project goes through a short preparation without AI (queued → processing, usually seconds) and gets the draft status. Wait for it by polling GET /v1/projects/{id}.

1. Read the video

   GET /v1/projects/{id}/transcript. Speech isn't recognized in a draft: fragments are by time (kind: "range"), and elements are placed by video time (Elements by time). Need captions? Recognise the speech (POST /v1/projects/{id}/transcribe, at cost per minute of sound): after that the captions follow the speech, and words with ids appear in the video text.

2. Build the video with operations

   POST /v1/projects/{id}/edits — split, trim, reorder and delete fragments, remove pauses (cut_silences), add a hook, titles, plaques, zooms, transitions and colour, set music and voice. Edits are free.

3. Export

   POST /v1/projects/{id}/export (Export and publishing). The first successful export moves the project to succeeded; if the export fails, the project stays a draft.

What's next#