Dynamic captions

Capability and preparation

Use subtitle to add dynamic captions or change their style, returning a new video while preserving the original. Requires vmake-labs-cli, targeting version 0.1.11, compatible with >=0.1.11 <0.2.0.

Before the first invocation, read CLI setup, resolve the executable, and check its version and the file parameters in subtitle --help. For unclear JSON fields or media limits, query the subtitle entry in vmake contracts. Read runtime and recovery for output, interactions, and errors.

• Caption content must come from recognizable speech or an available subtitle timeline. Obtain suitable input if neither is available.
• Use --subtitle-material-id for an actually configured valid style ID. Otherwise explain that a random style will be used and omit the option. Keep style IDs separate from session media IDs.
• For text edits, translation, highlighted words, or positioning, check current command support and available styles. Remove existing burned-in subtitles through the video subtitle-removal capability only when that fits the user's goal.
• Execute a clear caption request directly. Ask only when an ambiguous goal, such as "improve expression," leaves the intended operation unclear. Add montage or existing-caption removal only as needed for the request.

Collect input

1. Use the actual video file or URL provided. Do not repeat questions already answered; ask the user to choose if multiple candidates are ambiguous.
2. Use a readable local video. Download every remote original according to CLI setup, then submit the local file. If downloading fails, ask for the original local file and continue the established caption request when it arrives.
3. Add -r '<session_id>' when continuing a task; omit it for a new one. Pass optional --lang, --detach, and --subtitle-material-id only as supported by current help. Follow CLI setup for defaults and login.
4. When continuing a montage, retain its session and obtain the complete current Artifact or Message Part video URL. Download it first, then let the CLI upload it to its CDN. Query Task/history if the URL is missing, expired, or represented only by a media ID. Keep source-to-file mappings. Do not first use create-room, add-media, or --item-id.