| name | video-pipeline |
| description | Video production pipeline for creator brands — editing, quality review, intro/outro concat, publishing, and platform distribution. Use when user says "process video", "publish video", "video pipeline", "edit video", "quality review", "burn captions", "concat intro", "video auto-publish", or any video production activity. |
Video Pipeline Skill
Complete end-to-end video production workflow for a reusable media production team. This skill documents the canonical pipeline from raw recording to multi-platform publish.
Pipeline Order (ALWAYS follow this sequence)
- Edit Silence — Remove silent regions from raw video
- Transcribe — Generate transcript from audio
- Generate Captions — Create ASS/SRT subtitle file from transcript
- Burn Captions — Hard-code captions into video
- Quality Review — Automated quality check (MUST pass before proceeding)
- Concat Intro/Outro — Prepend/append brand bumpers
- Generate Variants — Create platform-specific aspect ratio versions
- Publish — Upload and schedule across all platforms
Configuration
All pipeline config lives in: data/content/video-pipeline/config.json
Key settings:
publishing.platforms — account IDs for each platformpublishing.profileId — Late/Zernio profile: {{LATE_PROFILE_ID}}introOutro.sourceDir — {{VIDEO_ASSET_DIR}}qualityReview.maxRetries — 2 attempts before escalating to the operator
Step 1: Silence Removal
Tool: vidpipe-remove_silence
Parameters:
videoPath: [input video path]
output: [output path with "-edited" suffix]
threshold: "-30dB"
minDuration: 0.5
Rules:
- Add 50ms padding on each side to avoid clipping words
- Output file naming:
{original-name}-edited.mp4
Step 2: Transcription
Use Gemini AI via analyze_video with prompt requesting word-level timestamps. Or use external transcription service. Output: JSON array of {start, end, text} segments.
Step 3: Caption Generation
Generate ASS subtitle file from transcript segments.
Caption Style Standards (default preference):
- Font: Arial Bold
- Size: 21pt (NEVER >24 for 720p vertical)
- Color: White text, black outline (2px), drop shadow
- Position: Bottom-center, MarginV=25+
- Max lines: 2 visible at once
- Word wrapping: Break at natural phrase boundaries
Platform-Specific Safe Zones:
| Platform | Aspect | Caption Zone |
|---|
| TikTok/Reels/Shorts | 9:16 | Middle 70% vertically |
| Instagram Feed | 1:1 | Bottom 30% |
| YouTube Main | 16:9 | Bottom 10% (traditional) |
Step 4: Burn Captions
Tool: vidpipe-burn_captions
Parameters:
videoPath: [edited video]
captionsFile: [ASS file path]
output: [output with "-captioned" suffix]
Step 5: Quality Review (CRITICAL — MUST PASS)
Tool: analyze_video (Gemini AI)
Prompt: [use qualityReview.promptTemplate from config.json]
Mandatory Checks:
- ✅ Captions stay in bottom 25% of frame (no overflow)
- ✅ Face clearly visible throughout (not obstructed)
- ✅ No bad editorial cuts (jump cuts, awkward transitions)
- ✅ Audio quality (no clipping, appropriate volume)
- ✅ Framing appropriate (face visible, not cut off)
Critical Lessons Learned (from data/content/video-pipeline/quality-checklist.md):
- FontSize=21 (not 36!), MarginV=25+ for 720×1280 vertical
- Face region (top 60% of vertical) must remain unobstructed
- After concat, verify output duration ≈ sum of inputs (±2s)
- Silence removal can clip word starts — use 50ms padding
Failure Protocol:
- FAIL → identify issue → fix → re-review (retry 1)
- Still FAIL → adjust approach → re-review (retry 2)
- Still FAIL after 2 retries → STOP. Notify the operator. Do NOT publish.
Log all new lessons to data/content/video-pipeline/quality-checklist.md.
Step 6: Intro/Outro Concat
Rules:
- Main videos: intro + outro
- Shorts: outro only (no intro)
- Medium clips: intro + outro
Aspect Ratio Matching:
| Video Ratio | Intro File | Outro File |
|---|
| 16:9 | {{VIDEO_ASSET_DIR}}\intro.mp4 | {{VIDEO_ASSET_DIR}}\outro.mp4 |
| 9:16 | {{VIDEO_ASSET_DIR}}\intro-portrait.mp4 | {{VIDEO_ASSET_DIR}}\outro-portrait.mp4 |
| 1:1 | {{VIDEO_ASSET_DIR}}\intro-square.mp4 | {{VIDEO_ASSET_DIR}}\outro-square.mp4 |
| 4:5 | {{VIDEO_ASSET_DIR}}\intro-feed.mp4 | {{VIDEO_ASSET_DIR}}\outro-feed.mp4 |
Concat Command (ALWAYS re-encode — never -c copy due to time_base mismatch):
ffmpeg -f concat -safe 0 -i filelist.txt -c:v libx264 -preset fast -crf 18 -c:a aac output.mp4
⚠️ CRITICAL: Caption-burned videos have time_base=1/3883000 from libx264, but bumpers have time_base=1/90000. Using -c copy produces multi-hour corrupted files. ALWAYS re-encode.
Post-concat validation: Check output duration ≈ sum of input durations (±2s). If wildly off → re-encode.
Step 7: Platform Variants
Tool: vidpipe-generate_variants
Parameters:
videoPath: [final concatenated video]
platforms: ["tiktok", "youtube", "instagram", "linkedin", "twitter"]
outputDir: [run output directory]
Step 8: Upload & Publish
Upload Flow
- Get presigned URL:
late_presign_upload(filename, content_type="video/mp4")
- PUT video to
uploadUrl
- Use
publicUrl in media_items when creating posts
Create Posts
late_create_post(
content: [platform-specific copy],
platforms: [{"platform": "youtube", "accountId": "{{YOUTUBE_ACCOUNT_ID}}"}, ...],
media_items: [{"type": "video", "url": "[publicUrl]"}],
queue_id: "{{LATE_PROFILE_ID}}",
timezone: "{{TIMEZONE}}"
)
Platform Account IDs
See late-publishing skill for the canonical account ID table. Key IDs for quick reference:
- Profile/Queue:
{{LATE_PROFILE_ID}}
- YouTube:
{{YOUTUBE_ACCOUNT_ID}}
Run ID & Output Structure
- Run ID format:
video-YYYY-MM-DD-NNN
- Output path:
data/content-editor-output/{run_id}/
- Context package:
data/content-editor-output/{run_id}/context-package.json
Retry Policy (from config)
| Stage | Max Retries | Failure Policy |
|---|
| Research | 2 | Continue with partial |
| Planning | 1 | Stop and notify |
| Editing | 2 | Stop and notify |
| Blog | 2 | Partial success (don't block video) |
| Social | 2 | Per-platform independence |
| Upload | 3 | Per-platform retry |
| Scheduling | 1 | Non-fatal |
Orchestration Model
- Orchestrator: content-editor
- Parallel lanes: editing, blog, social
- Sequential prereqs: research → planning
- Notify on complete: Always (notify the operator)
Anti-Patterns
- ❌ Using
-c copy for concat (time_base mismatch corruption)
- ❌ FontSize >24 for vertical video captions
- ❌ Publishing without quality review pass
- ❌ Blocking all platforms because one failed
- ❌ Generic "check out my video" social posts (must be specific to content)
- ❌ Missing cross-references to related articles and repos
