YouMind
Увійти

107 Hidden Gems: I Categorized and Open-Sourced 180 Video Skills for Claude

@GoSailGlobal
СПРОЩЕНА КИТАЙСЬКА27 вер. 2026 р.
283K
391
72
63
832

Коротко

A comprehensive catalog of 180 open-source video tools for AI agents, categorized into 10 groups and security-rated. Highlights 107 low-star 'hidden gems' and compares major frameworks like Remotion and HyperFrames.

180 repositories, 10 categories, every README read and security-rated.

This is a video tool list designed for coding agents, collecting open-source skills and toolkits that enable Claude Code, Codex, and similar agents to create, edit, and animate videos. The process cost about $200; categorization and ratings were handled by Jev reading each README individually. The result is a GitHub repository and an online page with filters, refreshed every 8 hours.

Previous posts tested what Jev could do; this one shows the actual output of using it.

Jason Zhu - inline image

I. Overview: Editing Dominates, Music Videos are Rare

  • Editing & Post-production: 35 repos, Top Star: FireRed-OpenStoryline (3.4k)
  • Explainer/Educational: 32 repos, Top Star: vox-director (2.0k)
  • Frameworks & General Toolkits: 29 repos, Top Star: OpenMontage (61.5k)
  • Product Demos & Promotions: 26 repos, Top Star: video-shotcraft (9.7k)
  • Short Videos & Talking Heads: 19 repos, Top Star: codex-storyboard (344)
  • Stories & Animation: 13 repos, Top Star: story-to-handdrawn-video (2.1k)
  • Motion Graphics & Logos: 11 repos, Top Star: diffusionstudio/lottie (5.5k)
  • Scripts & Learning: 9 repos, Top Star: screenwriting-skills (1.4k)
  • Digital Humans: 4 repos, Top Star: heygen-com/skills (454)
  • Music Videos: 2 repos, Top Star: PDoomVideo (1.1k)

Editing, explainers, frameworks, and product demos account for 122 repos, two-thirds of the total. This distribution signals that agent-driven video creation is currently most mature in processing existing footage and turning scripts into explainers, while digital humans and music videos remain early-stage.

Another number is more telling: 107 of the 180 repos have fewer than 50 stars, and only 20 exceed 1,000 stars. Most tools in this field are undiscovered; star-count rankings miss them entirely.

Jason Zhu - inline image

II. Four Hard Rules for Inclusion, Evaluated by Model Reading READMEs

The usability of a list depends on clear rules. These four rules apply:

  1. It must create/edit/animate video (3D web or prompt collections don't count).
  2. It must be for agents (skills, plugins, MCP servers, or agent-written toolkits).
  3. It must have a README (no README, no rating).
  4. For >50 stars, relevance is key; for <50 stars, quality checks apply (showcase output, one-command setup, clear deliverables, complete docs, min 5 stars).

These closed questions play to Jev's strengths. Previous tests showed it outperforms BERT-family models in zero-shot classification with high confidence. Its weakness is calculation (dates/counts), but judging "is this an agent video tool?" requires no computation.

Security ratings: 178 SAFE, 2 CAUTION. Ratings cover READMEs and install steps. Check before installing. Misclassifications can be reported via issues.

Jason Zhu - inline image

III. Framework Layer: Two Paths – HTML Rendering vs. Remotion Programming

To let agents make video, choose a base. The 29 framework repos split into two paths:

calesthio/OpenMontage (61.5k stars): An open-source agent video production system with 12 pipelines, 100+ tools, and 700+ skills. Start here for the big picture.

heygen-com/hyperframes (53.5k stars): Write HTML to render video, built specifically for agents. Many small repos build on this, especially paper-collage and info-animation styles.

The other path is Remotion, which uses React for programmatic video. 14 of the 26 product demo repos explicitly use it, compared to 5 for HyperFrames.

Choose based on control granularity. HyperFrames has a lower barrier (agents know HTML/CSS); Remotion offers stronger expressiveness (frame-perfect shots, transitions, timing). A niche gem: iart-ai/data-animation-skills (6 stars) turns CSVs into dynamic charts with frame-accurate numbers.

Jason Zhu - inline image

IV. Top Picks from the Four Major Categories

Editing & Post-production (35)

  • FireRedTeam/FireRed-OpenStoryline (3.4k): Director-style editing via natural language, LLM planning + tool orchestration, reusable style skills.
  • ayushozha/AdobePremiereProMCP (106): Premiere Pro MCP server with 1,027 tools covering timeline, color grading, mixing, effects, export. Lets agents take over pro software.
  • louisedesadeleer/cut-video (101): Trims long recordings, removing silences/fillers but keeping laughter/comedic pauses. Few tools get this trade-off right.

Mainstream approach: ffmpeg/Whisper/vision models for rough cuts, then Remotion/HyperFrames for subtitles/motion. Talking-head videos are the most common scenario.

Explainer/Educational (32)

  • Alisa0808/vox-director (2.0k): Turns topics into Vox-style paper-collage explainers.
  • Vincentwei1021/video-talkcraft (1.2k): Voiceover-driven explainers with word-synced audio and 109 motion recipe cards.
  • aijiduonadegou/Paper-Cut (16): Paper-collage science without video models. Uses image models for final art, splits layers, animates via code. Saves the most expensive step.

Visual styles cluster around Vox paper-collage and 3Blue1Brown-style math animations.

Product Demos & Promotions (26)

  • Vincentwei1021/video-shotcraft (9.7k): Cinematic product videos via Remotion, with 152 shot recipe cards and 209 motion previews. These cards are reused by many repos.
  • op7418/guizang-product-video-skill (445): Reuses real product components/design language for update promos. Visuals match the product exactly.
  • Arman-Luthra/aftr (22): After Effects version of Puppeteer. The only AE-path repo in this category.

Short Videos & Talking Heads (19)

  • Yuuhann1999/codex-storyboard (344): Local multi-project storyboard workbench, managing everything from storyboard to render.
  • hassancs91/claude-faceless-shorts-creator (268): Faceless YouTube Shorts factory. Remotion visuals + ElevenLabs voice + word-synced captions. Full pipeline.
  • pyang5166/gbro-collage-info (51): Halftone paper-collage info animations generated from narration scripts, no image model dependency.
Jason Zhu - inline image

V. Small Categories Hide the Most Interesting Tools

Smaller categories often have the strongest personality.

Motion Graphics & Logos: diffusionstudio/lottie (5.5k) lets agents generate production-ready Lottie animations. nolangz/pixel2motion (2.3k) converts bitmap logos to SVG animations with QA evidence.

Stories & Animation: Eastern themes dominate. gnipbao/story-to-handdrawn-video (2.1k) turns Chinese stories into hand-drawn diary comic animations. AllenAI2014/remotion-guofeng-starter (25) creates Guofeng paper-cut animations from poems or idioms.

Scripts & Learning: Handles pre-production. jtydhr88/screenwriting-skills (1.4k) packages 26 professional screenwriting/dramaturgy skills. eternityspring/reelbench-skills (843) does shot-by-shot analysis, synthesizing videos with storyboard info.

Digital Humans: Only 4 repos, 3 based on HeyGen. Differences lie in encapsulation levels (generation only vs. social publishing).

Music Videos: Only 2, but both worth inspecting. JohnHeibel/PDoomVideo (1.1k) is the full source for the MV "I'm Upping My P(doom)". ledbetterljoshua/functional-emotions-video (45) features a custom GPU stroke renderer with 7 parallel chapter agents.

VI. Three Observations After Compilation

  1. Code-based rendering is dominant. 46 repos mention Remotion, 26 mention HyperFrames. Together they account for 40% of the list. Reason: Code rendering offers frame-level control, reproducibility, and modifiability—tasks agents excel at.
  2. Recipe cards are becoming universal assets. video-shotcraft's 152 shot cards and video-talkcraft's 109 motion cards do the same thing: break good shots into parameters agents can directly apply. More stable than long prompts.
  3. Small repos are severely underestimated. Among the 107 sub-50-star repos are gems like Paper-Cut (skips video models) and aftr (bridges AE). Star-count sorting misses them; rule-based filtering finds them.

Both the repo and online page are open-sourced. Recommend new repos via issues with links; they undergo the same review as existing entries:

https://github.com/zhuyansen/awesome-claude-video-skills

https://agentskillshub.top/best/claude-video-skills/

Creating content isn't easy. If you're interested in going global, follow WeChat public account 'Jason AI Going Global'. Join the GoSail community to discuss digital nomads, globalization, and AI.

Переробити в YouMind

Перетворіть одну віральну статтю на повноцінний робочий процес

Збирайте джерела, розшифровуйте патерни, створюйте матеріали, пишіть чернетки та поширюйте контент в одному AI-робочому просторі.

Дослідити YouMind
Для авторів

Перетворіть свій Markdown на охайну статтю для 𝕏

Коли ви публікуєте власні лонгріди, зображення, таблиці та блоки коду роблять форматування в 𝕏 складним. YouMind перетворює повну чернетку в Markdown на чисту статтю для 𝕏, готову до публікації.

Спробувати Markdown для 𝕏

Більше патернів для аналізу

Останні віральні статті

Переглянути більше віральних статей