article2media Multiple formats from articles and video

article2media is a collection of Claude Code/Codex skills that turns articles and audiovisual sources into visual posts or explainer videos.

Capabilities exchange files in a separate working directory, keeping text processing separate from visual layout. Original-text layout can use an article file directly; rewriting and translation require the agent to prepare inputs, since rendering commands do not perform that semantic work automatically.

The author's X article provides design background and examples; commands, counts and limits here follow the pinned repository and source code. The code uses MIT, while rights to sample articles and media remain subject to their original source notices.

The Claude Code plugin has the documented entry points below. After installation, use /article2media for the entry skill and namespaced capability names such as /article2media:image-cards.

The following skills CLI command installs for both Claude Code and Codex at user scope.

Codex uses $skill names in conversation, such as $article2media for the entry skill or $wechat-layout for one capability. Standalone skill names differ from the Claude plugin namespace; commands prefixed with article2media: do not work as universal command names across hosts.

Without an explicit route, long images and cards default to rewriting, while WeChat defaults to original-text layout. To preserve wording, state "original-text layout, preserve the body" in the task; narration, whiteboards and comics each use their own scripts or ledgers and do not carry a full-text verbatim guarantee.

Translation defaults to Simplified Chinese, and Chinese cleanup in video transcription also converts Traditional to Simplified. For a Traditional Chinese workflow, explicitly ask the agent to deliver Traditional Chinese, retain the original subtitles or transcript/raw.txt, and check the final text; --lang zh selects the speech-recognition language and does not preserve Traditional characters.

The official environment check below runs from the repository root. It does not install Node, Chrome, ffmpeg or models for you, but doctor starts temporary headless Chrome, tests Google Fonts connectivity and probes external tools.

Chrome uses a new temporary profile without automatically inheriting everyday browser login state. When a video source requires cookies, they go to yt-dlp only through explicit use of --cookies-from-browser; existing local audiovisual or subtitle files are another option.

--no-proxy skips only the two general-page proxy routes and is not a global offline switch. Some YouTube yt-dlp routes may fetch remote EJS components; image generation, external narration, fonts and source images each have their own network requests.

The pinned version contains 21 theme CSS files. Render the same content.json with another theme without rewriting the article for a color change, then check the layout again.

Some fonts load from Google Fonts or jsDelivr, so offline access or missing fonts may trigger fallbacks. Inspect the actual fonts recorded in QA and the output's line breaks instead of assuming identical layout from the theme name.

The agent first writes video/video.json to describe the audience, shots, lines and visuals, then runs video commands. Expected output is a 1080×1920 vertical video with covers, captions, review data and a publishing draft.

Run these commands from the repository root. Both make and render may synthesize narration when the cache is missing, and optional illustrations may call an image service; render is not guaranteed to replay without network access or costs.

The pinned catalog has 6 public voices, with news-man as the default; the 7 voices in the author's X article do not apply to this pinned version. Validate voice quality and speed on the target host.

Comic character-reference images currently pass through the Codex image route; do not assume every provider uses the same mechanism. Whiteboards need the source inventory retained to check where each original paragraph went and its source hash.

Prepare Traditional Chinese Markdown you have the right to use, with a title, paragraphs, subheadings and a list, and save it as owned-notes.md in the current directory. This example uses the original-text route to deliver 3:4 cards and quiet-themed WeChat layout, without model-generated illustrations or narration.

Install all skills with the earlier skills CLI command, then run these commands from the directory containing owned-notes.md. They use the global Codex skills root; Claude plugin users must use their actual plugin installation location instead of assuming the same path.

This is a suggested procedure; this research did not execute the commands or publish content. For a summary, have the agent write content.json first; for translation, write and structurally check the translation, review it manually, then pass the appropriate file to the layout capability.

See the implementations in render.mjs and the video review code. This guide records the pinned version's behavior without modifying upstream or labeling unexecuted examples as passes.

When researching or changing the engine, run official self-tests in a separate repository. They generate files and perform the corresponding rendering, requiring a browser and prepared environment; this research did not run these upstream tests.

Change source content through article.md/content.json and layouts through layout/theme; manual edits to generated HTML are not reproducible fixes. Retain program reports and previews and explain any unfinished checks.

When reporting an issue, provide the following and remove credentials and private content unrelated to the problem.

• Commit.

• Machine.

• Agent host.

• Input route.

• Commands run.

• Relevant QA files.

Diagnose incomplete sources, model output and rendered layout separately.

Use owned Markdown to complete original-text cards or WeChat layout and check both files and the actual platform preview. Add the corresponding agent work, models and tools when you need rewriting, translation, video or comics.

• Official README: installation, themes and output entry points.

• Shared pipeline: host names, default routes and handoff files.

• Original-text acceptance specification: text-fidelity scope and rerun rules.

• Issues: publicly reproducible problems and QA information.