Capcut Draft

CapCut Draft Manipulation (剪映草稿操控)

This skill enables Claude to directly manipulate CapCut/JianYing (剪映) draft files on the local filesystem using built-in tools (Read, Write, Edit, Bash). No external scripts or servers needed.

Supported Operations

• List all drafts with metadata
• Create new drafts with video clips, subtitles, audio, effects, transitions, animations
• Read draft structure and content
• Modify existing drafts (add/remove/adjust segments, subtitles, effects)
• Delete drafts

Draft Root Discovery (草稿根目录)

CapCut stores all drafts under a root directory. Find it by checking these paths in order:

macOS:

~/Movies/JianyingPro/User Data/Projects/com.lveditor.draft

If not found, try:

~/Library/Containers/com.lemon.lvpro/Data/Movies/JianyingPro/User Data/Projects/com.lveditor.draft

Windows:

%LOCALAPPDATA%\JianyingPro\User Data\Projects\com.lveditor.draft

To discover programmatically:

# macOS
ls ~/Movies/JianyingPro/User\ Data/Projects/com.lveditor.draft/root_meta_info.json 2>/dev/null && echo "FOUND"

# Windows (Git Bash)
ls "$LOCALAPPDATA/JianyingPro/User Data/Projects/com.lveditor.draft/root_meta_info.json" 2>/dev/null && echo "FOUND"

If the user provides a custom draft root path, use that instead.

File Structure

<DraftRoot>/
├── root_meta_info.json          # Global index of all drafts
├── MyDraft/                     # One folder per draft
│   ├── draft_info.json          # Main content (tracks, materials, segments)
│   ├── draft_meta_info.json     # Metadata
│   └── draft_cover.jpg          # Cover image
└── AnotherDraft/
    └── ...

Time System (时间系统)

CapCut uses microseconds for all time values.

• Convert seconds to microseconds: seconds × 1,000,000
• Convert microseconds to seconds: microseconds ÷ 1,000,000
• Default draft duration: 40,000,000 (40 seconds)
• Animation duration: ~`500,000` (0.5 seconds)
• Transition duration: ~`466,666` (0.47 seconds)

UUID Rules

All IDs must be uppercase UUIDs. Generate with:

uuidgen | tr '[:lower:]' '[:upper:]'
# Example output: A1B2C3D4-E5F6-7890-ABCD-EF1234567890

When using effect/transition presets, always regenerate the id field with a new UUID. Only reuse effect_id and resource_id.

Canvas Presets (画布预设)

Render Index Values

Quick Operations

List all drafts

Read root_meta_info.json → iterate all_draft_store → show table with draft_name, creation time, duration, draft_id.

Create a draft

See references/operations.md for the full workflow. Summary:

1. Discover root → generate UUID → create folder
2. Build draft_info.json with canvas_config, config, materials, tracks
3. Write draft_meta_info.json
4. Copy cover image
5. Update root_meta_info.json (prepend to all_draft_store, increment draft_ids)

Modify a draft

Read draft_info.json → make changes → write back. Common modifications:

• Add video segment: new Video material + Canvas + Speed + SoundProjectMapping + TrackSegment on video track
• Add subtitle: new Text material + TrackSegment on text track (render_index: 14002, clip.transform.y: -0.75)
• Add effect: clone preset with new UUID + TrackSegment on effect track (render_index: 11005)
• Adjust timing: modify target_timerange on segment

Delete a draft

Remove entry from root_meta_info.json → delete draft folder with rm -rf

Key Defaults

• Text font size: 8.0
• Text color: #FFFFFF
• Text Y position: -0.75 (bottom of screen)
• Text content format: <size=8.0>[text content]</size>
• Audio volume: 0.3162277638912201
• Video clip scale: {x: 2, y: 2}
• Video HDR: {intensity: 1.0, mode: 1, nits: 1000}
• version: 360000
• new_version: "79.0.0"
• source: "default"

Short-Video Captions (短视频字幕)

For livestream-sliced short videos (抖音/快手 9:16), subtitles must stay readable on mobile over busy backgrounds. Apply this style to every material_text/segment_text you emit for a live-slicer clip draft, overriding the generic text defaults above:

Chunking: ASR sentences are often 10-20 chars — too long for one line. Split a sentence >8 chars into ≤2 chunks at a natural pause (comma, conjunction, breath) and create a separate text segment per chunk. Distribute each chunk's duration by character-count proportion from the sentence's total time (chunk.duration = sentence.duration × chunk.chars / sentence.chars) so chunks tile the sentence without overlap and with no gaps. Word-level exact alignment is out of scope for now; this proportional split is the approximation.

Important Notes

• Close CapCut before modifying drafts — if CapCut is running, it may overwrite your changes or cause corruption
• Always validate JSON after writing — invalid JSON will crash CapCut
• material_id references must be consistent — every segment's material_id must exist in the corresponding materials array
• extra_material_refs must resolve — video segments reference audio, canvas, speed, and animation IDs; all must exist
• No time overlap within the same track — segments on the same track must not have overlapping target_timerange

JSON Templates (JSON 模板)

All templates use __PLACEHOLDER__ format. Copy a template, replace placeholders with actual values.

Start here:

• templates/example_minimal.json — Complete working example (1 video + 1 text + 1 audio). All IDs correctly linked. Use as primary reference.

Complete draft files:

• templates/full_draft_info.json — Full draft_info.json with all sections (copy → fill per-segment arrays)
• templates/full_draft_meta_info.json — draft_meta_info.json (5 unique placeholders)
• templates/root_meta_entry.json — Entry to prepend into root_meta_info.json's all_draft_store (5 unique placeholders)

Material templates (add to materials.*[] arrays):

• templates/material_video.json — Video/photo material (6 placeholders: VIDEO_ID, FILE_PATH, FILE_NAME, IMAGE_WIDTH, IMAGE_HEIGHT, SEGMENT_DURATION_US)
• templates/material_audio.json — Audio material (4 placeholders: AUDIO_ID, AUDIO_DURATION_US, AUDIO_FILE_NAME, AUDIO_FILE_PATH)
• templates/material_text.json — Text/subtitle material (2 placeholders: TEXT_ID, TEXT_CONTENT)
• templates/material_effect.json — Video effect (clone from preset, replace instance ID + preset IDs)
• templates/material_transition.json — Transition (clone from preset, replace instance ID + preset IDs)
• templates/material_animation.json — Animation container (fill from presets)

Track segment templates (add to track.segments[] arrays):

• templates/segment_video.json — Video segment (references VIDEO_ID, AUDIO_ID, CANVAS_ID, SPEED_ID, timing)
• templates/segment_audio.json — Audio segment (references AUDIO_ID, duration)
• templates/segment_text.json — Text segment (references TEXT_ID, timing; render_index=14002, y=-0.75)
• templates/segment_effect.json — Effect segment (references EFFECT_ID, timing; render_index=11005)

Auxiliary material templates (one per video segment):

• templates/aux_canvas.json — Canvas material (1 placeholder: CANVAS_ID)
• templates/aux_speed.json — Speed material (1 placeholder: SPEED_ID)
• templates/aux_sound_project.json — Sound project mapping (1 placeholder: SOUND_PROJECT_ID)

Workflow: Create a new draft using templates

1. Generate UUIDs: for i in $(seq 1 20); do uuidgen | tr '[:lower:]' '[:upper:]'; done
2. Copy full_draft_info.json → replace top-level placeholders (CANVAS, FPS, DRAFT_ID, DURATION)
3. For each video segment: copy material_video.json + segment_video.json + aux_canvas.json + aux_speed.json + aux_sound_project.json, fill IDs/timing, insert into the arrays
4. For each subtitle: copy material_text.json + segment_text.json, fill TEXT_ID/timing
5. For audio: copy material_audio.json + segment_audio.json, fill AUDIO_ID/duration
6. Copy full_draft_meta_info.json → replace placeholders
7. Copy root_meta_entry.json → prepend to root_meta_info.json's all_draft_store

Reference Files

For detailed specifications, read these files:

• references/draft-structure.md — Complete JSON schema for all three draft files
• references/presets.md — All animation, effect, and transition presets with exact IDs and parameters
• references/operations.md — Detailed step-by-step guides for each CRUD operation