Clip-to-Post Skill

中文说明

Clip-to-Post is a reusable AI workflow skill for turning visual material into publish-ready social content: storyboard panels, captions, cover art, and exportable workflow state.

It is not a "one prompt does everything" trick. It distills a more reliable creative pipeline:

video / images
  -> keyframes
  -> step analysis
  -> base storyboard
  -> character + watermark integration
  -> refined panels
  -> social captions
  -> cover art
  -> export package

Why It Matters

Most AI content tools collapse everything into one giant model call. That is hard to debug, hard to steer, and hard to productize.

Clip-to-Post treats content generation like a small production line:

• Understand before generating: analyze keyframes into semantic steps first.
• Draft before polishing: generate the whole storyboard, then refine panels one by one.
• Write from final visuals: generate captions after the images exist.
• Design for platforms: use strategy prompts for Xiaohongshu, Instagram, SOP docs, and custom channels.
• Keep humans in the loop: allow editing steps, locking panels, selecting captions, and rerunning only stale nodes.

Best For

• AI content factories
• Short-video repurposing tools
• Multimodal workflow demos
• Agent / Skill marketplaces
• Storyboard generation
• Xiaohongshu and Instagram post generation
• Product tutorials and internal SOP visual guides

What Is Inside

.
├── SKILL.md
├── README.md
├── README.zh-CN.md
├── assets/
│   └── clip-to-post-hero.png
├── scripts/
│   └── extract_frames.py
└── references/
    ├── pipeline-blueprint.md
    ├── prompt-templates.md
    └── schemas.md

Frame Extraction Reality Check

This skill now includes a local video frame extraction helper, but it still does not magically play every video on the internet.

Source	Supported by this repo	Notes
Local .mp4, .mov, .mkv files	Yes, via scripts/extract_frames.py	Requires ffmpeg and ffprobe
Ordered image batches / screenshots	Yes	Use images directly as FrameData[]
Browser video already loaded in a host app	Contract only	Host app should use <video> + canvas.drawImage()
Bilibili, Instagram, logged-in pages, expiring URLs, DRM/CORS media	Not directly	Requires host browser/backend extraction, proxy, permissions, or platform-specific tooling

SKILL.md

The main skill entrypoint. It defines when to use the skill and how to execute the segmented multimodal workflow.

references/pipeline-blueprint.md

The ClipSketch-inspired pipeline architecture: node responsibilities, provider abstraction, state persistence, export behavior, and failure handling.

references/prompt-templates.md

Reusable prompt templates for step analysis, storyboard generation, character integration, panel refinement, caption generation, and cover generation.

references/schemas.md

TypeScript-style contracts for input, frame data, step analysis, artifacts, pipeline state, provider interfaces, and export payloads.

Pipeline

flowchart LR
  A["Video / Image Batch"] --> B["Keyframe Capture"]
  B --> C["Step Analysis"]
  C --> D["Base Storyboard"]
  D --> E["Character / Watermark"]
  E --> F["Panel Refinement"]
  F --> G["Caption Generation"]
  G --> H["Caption Selection"]
  H --> I["Cover Generation"]
  I --> J["Export Package"]

Loading

Quick Start

Clone the repository:

git clone https://github.com/501981732/clip-to-post.git
cd clip-to-post

Use the skill folder directly in a Codex/Claude-compatible skill environment, or package it from a clean checkout.

Validate with the skill creator tooling:

python3 /path/to/skill-creator/scripts/quick_validate.py /path/to/clip-to-post

Create a clean distributable zip from the git checkout:

mkdir -p dist
git archive --format=zip --prefix=clip-to-post/ HEAD -o dist/clip-to-post.zip

Extract local video frames:

python3 scripts/extract_frames.py ./demo.mp4 --timestamps 0,2.5,5 --out ./frames --include-base64

Or sample at intervals:

python3 scripts/extract_frames.py ./demo.mp4 --every-seconds 2 --max-frames 8 --out ./frames

The script writes JPEG frames and manifest.json. Use the manifest frames as the visual input for the rest of the pipeline.

Example Trigger Prompts

Use clip-to-post to turn these cooking video frames into a Xiaohongshu post package.

Design an AI workflow that converts product tutorial screenshots into storyboard panels, captions, and cover art.

Repurpose this travel video into Instagram carousel panels with three caption options.

Use this as an internal SOP generator: analyze screenshots, produce visual steps, and export an implementation-ready state schema.

Design Principles

• Small AI nodes beat giant prompts.
• Intermediate artifacts should be visible, editable, and persisted.
• Platform strategy belongs in a strategy layer, not scattered across prompts.
• Generated images and captions should share the same source truth.
• Reruns should invalidate only downstream artifacts.

Roadmap Ideas

• Add more platform strategies: Douyin, WeChat, TikTok, YouTube Shorts.
• Add domain strategies: baby memory, food recipe, fitness movement, auto review, product tutorial, travel log, internal SOP.
• Add a sample workflow runner.
• Add exported demo packages.
• Add cost and model-routing reference docs.
• Add browser extraction adapters for host apps that can legally access remote video playback.

Keywords

ai-skill multimodal-ai agent-workflow video-to-post storyboard-generation content-repurposing prompt-engineering human-in-the-loop xiaohongshu instagram