55 Open-Source AI Video Skills Architecture and Automated Pipeline Guide

1. Topic Selection and Planning (4 Skills)

The initial step in video production is determining the core topic and creating an actionable production plan across 4 steps.

• ra-选题: Topic lifecycle management Skill. Creates an Obsidian topic card upon receiving inspiration prompts; queries reference libraries and 3 day methodologies upon deep-dive requests; fetches competitor titles and 3 day recent works sorted by traffic potential upon recommendation requests to output 5 candidate lists. Automatically routes approved topics to the video workspace.
• ra-实操策划: Long-form video and practical tutorial planning Skill. Generates detailed planning scripts for 8 to 15 minute Bilibili and YouTube videos, incorporating test prompt suites, structural timelines (annotating talking-head vs. screen-share segments), voiceover scripts, and recording checklists, managed via frontmatter status fields.
• ra-hook: Short video opening hook selection Skill for 5 second short videos. Categorizes hooks into 7 sub-types across 3 dimensions (content suspense, audience pain points, event trends), providing 3 prerequisites, sentence templates, and common anti-patterns for 5 second hook matching.
• ra-video-title: Video title generation Skill. Locks the one-sentence core theme, analyzes competitor title patterns, and generates 8 to 12 two-part candidate titles with Top 3 recommendations.

2. Content Creation (5 Skills)

Transforming raw reference materials into standardized, production-ready 5 video scripts.

• ra-video-wash-pipeline: Script washing orchestration Skill covering 5 stages. Accepts video URLs or local files and automatically orchestrates video downloading, transcript extraction, script rewriting, quality inspection, and production queueing.
• ra-逐字稿提取skill: Watermark removal and ASR transcription 3 step Skill. Fetches raw video via watermark removal APIs and extracts verbatim transcripts with precise timestamps using Paraformer ASR. Transcripts remain internal intermediates in privacy storage without exposing source traces in final outputs.
• ra-洗稿: Script rewriting master Skill. Reconstructs transcripts into fresh video scripts by chaining 5 modules: ra-人话 for AI de-patterning, dbs-ai-check for machine phrase scanning, dbs-hook for first 5 second opening engagement, dbs-resonate for emotional resonance, and ra-video-title for candidate titles to generate 8 to 12 titles before outputting standard handoff drafts.
• ra-人话: De-AI writing and oral expression refactoring Skill. Enforces strict rules against binary contrasting formulas, pseudo-insight markers ("essentially", "truly"), lecture-style colons, and vague focus shifts. All public-facing Chinese scripts must pass through this Skill.
• ra-公众号提取: WeChat Official Account article extraction Skill. Bypasses access restrictions using MicroMessenger UA strings to extract plain text without requiring API keys, feeding raw text into downstream washing and topic pipelines across 5 items.

3. Video Downloading (2 Skills)

Providing efficient and compliant local fetching capabilities for cross-platform 2 video and audio assets.

• ra-video-download: Multi-platform video downloading Skill. Supports Douyin, YouTube, Bilibili, Twitter/X, and RedNote video fetching via yt-dlp, prioritizing TikHub APIs for watermark-free Douyin source files. Downloaded assets reside strictly in local privacy directories .internal/.
• xiaohu-video-download: Overseas and batch video processing Skill. Extends baseline fetching with audio isolation, playlist batch downloading, and local subtitle hardcoding, tailored for overseas video translation and asset re-purposing.

4. Voiceover Synthesis (1 Skill)

Establishing a standardized local voice cloning mechanism for 1 local voice synthesis.

• tts-skill: Local IndexTTS2 voice cloning Skill. Uses fixed lossless WAV audio references for calm tone synthesis, applying atempo=1.12 pitch-preserved speed adjustment. Generates voice_manifest.json tracking parameters and reference hashes, explicitly prohibiting cloud TTS fallback.

5. Digital Avatar Generation (1 Skill)

Enabling automated digital avatar video generation and synthesis parameter control for 1 avatar mechanism.

• heygen-digital-avatar: HeyGen digital avatar synthesis Skill. Calls a pre-trained personal Digital Twin avatar (black T-shirt seated pose) rendered as a circular crop in the bottom-left corner. Manages OAuth credentials, avatar selection, and synthesis. Enforces paid generation gates requiring human-approved audio previews.

6. Video Editing (3 Skills)

Providing multi-scenario editing capabilities from local talking-head rough cuts to multi-track editing and FCPXML export for 3 editing capabilities.

• ra-local-talking-head-cut: Local automated talking-head rough-cut Skill. Generates proofreading SRTs via Volcengine ASR, pauses for human confirmation, and executes semantic editing (removing repeated takes), compressing pauses to 380 to 450ms, normalizing loudness to -16 LUFS, using up to 3 restrained transitions, and exporting QC-verified MP4 files.
• video-use: General-purpose video editing Skill. Provides end-to-end video processing including transcription, trimming, color grading, motion overlays, and subtitle hardcoding, suitable for non-talking-head footage and multi-take selection.
• AI剪口播: Speech error identification and XML export Skill. Identifies speech blunders via Volcengine ASR, generating deletion manifests and exporting FCPXML files for fine-tuning in CapCut and Final Cut Pro.

7. Subtitle Engineering (2 Skills)

Decoupling subtitle timestamp extraction from visual styling and rendering for 2 modules.

• ra-audio-to-subtitles: Word-level timestamp extraction and QC Skill. Extracts word-level timestamps via Doubao-ASR to produce JSON and SRT captions. Automatically inspects broken English terms, isolated words, cross-screen conjunctions, and reading speeds, generating caption-qc.json reports.
• skill-captions: Subtitle styling and hardcoding Skill. Applies the anchor-dark default style (STHeiti Medium font, charcoal semi-transparent rounded background, white text, fixed bottom anchor, 2x 4K redraw). Renders preview frames for confirmation prior to full video hardcoding.

Subtitle Pipeline Division of Labor

• ra-audio-to-subtitles (Timestamps & QC)

  Extracts word-level timestamps and automatically scans for broken terminology, isolated words, and reading speed issues to produce JSON manifests.

• skill-captions (Styling & Hardcoding)

  Applies anchor-dark visual styling, generates representative preview frames, and executes 4K native redraw and video hardcoding upon approval.

8. Visuals and Cover Design (6 Skills)

Comprehensive visual asset generation spanning IP illustrations, cover designs, and motion collages across 6 visual assets.

• ian-xiaohei-cat-illustrations: Xiaohei Cat IP illustration Skill. Generates white hand-drawn style concept illustrations with red/orange/blue annotations using DashScope Wanx, keeping images text-free for dynamic text overlay during rendering.
• ian-xiaohei-illustrations: General concept visualization Skill for 5 scenarios. Inherits the Xiaohei aesthetic while expanding beyond the cat mascot for broader technical concept illustration across 5 scenarios.
• skill-cover: Cover preset management Skill. Registers multiple cover design styles and dual aspect ratio (16 to 9 and 3 to 4) video cover asset registration.
• editorial-dot-cover: Editorial dot-style cover Skill. Combines warm gray paper textures, bold black typography, generous white space, and dot vector icons to output editable SVG and PNG files.
• editorial-collage-motion: Halftone paper collage motion Skill. Deconstructs reference images into collage rules, generating static layers before compiling step-by-step assembly animations in FFmpeg or HyperFrames.
• rn-cover-skill: Editorial diagram cover Skill. Generates 5 to 2 warm-white canvas covers directly from titles and summaries, placing large typography on the left and real-time concept graphics on the right.

9. Graphic Content Creation (1 Skill)

Converting long-form Markdown articles into social-media-ready image sets for 1 graphic module.

• xhs-article-to-images: RedNote graphic conversion Skill. Parses long-form Markdown into 3 to 4 HTML layouts across 5 design themes (Elegance, Warm Latte, Forest Pine, Obsidian Gold, Night Reader) and 3 topic styles, outputting high-resolution PNG image sets via Playwright screenshots.

10. Production Direction and QA (2 Skills)

Central control engine for orchestrating automated video production and post-release reviews across 2 direction systems.

• ra-video-production-director: Production director Skill. Reads handoff draft frontmatter contracts (aspect ratio, duration, voice, subtitle style), dispatches downstream Skills, manages folder state transitions, and enforces pre-archive rendering QA.
• ra-复盘: Analytics and asset reflection Skill. Fetches post-release performance data, categorizes viral videos into R-tier and M-tier, attributes success factors, and archives key insights back into topic cards.

11. HyperFrames Motion Graphics (6 Skills)

Code-driven video motion graphics architecture built on the HyperFrames framework across 6 motion graphics skills.

• rn-motion-director: Motion graphics director Skill. Translates text scripts into visual metaphors, motion syntax, and beat graphs, enforcing Anti-PPT quality audits to ensure continuous motion.
• rn-motion-replica: Motion reference replication Skill. Deconstructs reference video layouts timing and transitions to construct original, editable HyperFrames code projects.
• rn-dark-saas-video: Dark SaaS product video Skill. Features dark starry stages, purple ambient lighting, animated typography, and gradient CTAs across 8 scene blueprints and 3 duration presets.
• rn-bw-text-opener: Black-and-white typewriter opener Skill. Delivers pure black backgrounds, white typewriter text animations, and synchronized audio effects with 3 duration presets and Python timing scripts.
• rn-replica-qc: Replication quality control Skill. Executes 5 level fidelity assessments across asset, runtime, and delivery 3 full-frame QA gates.

12. dbs Business Diagnostic Toolkit (22 Skills)

Integrated from the open-source dbskill project by @dontbesilent, serving as a core suite for business diagnostics, content QA, and strategy formulation across 22 core tools.

dbs Toolkit Invocation Modes

• Automated Pipeline Invocation

  Triggered automatically within video washing and topic chains, focusing on AI pattern removal, first 5 second opening hook QA, and emotional resonance.

• Interactive Independent Invocation

  Explicitly executed by developers for root-cause business model analysis, competitor benchmarking, execution unblocking, and diagnostic reporting.

Based on execution triggers within the video production line, dbs Skills fall into automated invocation and independent diagnostic categories:

1. Automated Pipeline Invocations across 3 dbs Skills:

   • dbs-ai-check: Text AI pattern scanner Skill. Detects mechanical phrasing and flat logic to assist ra-洗稿 in refining natural phrasing.
   • dbs-hook: Opening hook diagnostic Skill. Evaluates short video first 5 second engagement and provides actionable revision prompts.
   • dbs-resonate: Resonance diagnostic Skill. Verifies whether scripts address audience pain points, preventing comprehensive yet bland content.

2. Independent Business Diagnostics across 7 dbs Skills:

   • dbs-diagnosis: Business model root-cause diagnostics. Analyzes underlying commercial logic, pricing strategies, and customer fit.
   • dbs-content: Content strategy and topic QA. Evaluates macro topic direction and long-tail content value.
   • dbs-benchmark: Competitor benchmark filtering. Locates high-quality reference accounts using 5 stage filtering rules.
   • dbs-unblock: Execution unblocking diagnostics. Applies Adlerian psychology frameworks to identify operational bottlenecks and resolution paths.
   • dbs-deconstruct: Business concept deconstruction. Uses linguistic philosophy to break down vague commercial terms into atomic components.
   • dbs-save / dbs-restore: Session state archiving and restoration. Saves diagnostic contexts as Markdown snapshots for cross-session continuity.
   • dbs-report: Diagnostic report generation. Consolidates multi-snapshot sessions into structured business review reports.
   • Extended Business Strategy Modules: Includes 12 specialized diagnostic and derivative tools (such as dbs-persona, dbs-pricing, dbs-pitch, dbs-traffic, dbs-review).

Skill Orchestration Network and Chain Execution

Most Skills chain automatically via explicit contracts, enabling multi-stage automated execution from a single command.

Automated Script Washing Pipeline Chain

1. Fetching and Transcription

   ra-video-wash-pipeline dispatches ra-video-download to fetch source videos, then triggers ra-逐字稿提取skill for timestamped transcripts.

2. Script Rewriting and De-AI Refining

   Triggers ra-洗稿 to rewrite scripts, automatically calling ra-人话 to remove AI writing patterns.

3. Quality QA and Title Generation

   Dispatches dbs-ai-check, dbs-hook, and dbs-resonate for QA scanning, finally calling ra-video-title to queue candidate titles.

Video Production and Rendering Chain

1. Voiceover and Avatar Synthesis

   ra-video-production-director validates contracts, dispatches tts-skill for audio, and calls heygen-digital-avatar upon confirmation.

2. Motion Rendering and Subtitle Hardcoding

   Calls ian-xiaohei-cat-illustrations for artwork, renders HyperFrames motion graphics, extracts timestamps via ra-audio-to-subtitles, and hardcodes subtitles using skill-captions.

3. QA and Automated Archiving

   Executes full visual and media stream QA, automatically archiving completed projects upon passing all checks.

Beyond primary pipelines, hook chains (ra-hook selection and dbs-hook polishing) and cover chains (skill-cover candidate generation and human approval) operate as modular, decoupled workflows.

Zero-Code Orchestration: Rapid Production Implementation

Encapsulating complex procedures into Skills and establishing automated dispatching drastically reduces human effort.

Rapid Video Production Operational Workflow

1. Trigger Automated Script Washing

   Send command to AI to make video from URL. ra-video-wash-pipeline automatically chains 8 Skills and queues handoff drafts within minutes.

2. Automated Background Rendering

   Codex retrieves queued tasks automatically while ra-video-production-director orchestrates voiceover, avatar synthesis, subtitles, and QA.

3. Lightweight Human Checkpoint Approval

   Human creators approve key decisions across 3 checkpoints: reviewing proofreading drafts, approving audio previews, and selecting cover designs.

Empirical testing confirms that all 3 human checkpoints require only 5 minutes combined. Producing 30 videos per month requires just 150 minutes of direct interaction, keeping total effort under 10 hours monthly when including recording and initial planning.

Complete 55 Open-Source Skills Reference Table