A public Codex Skill for turning long-form Chinese articles into polished vertical short videos.
中文一句话：这是一个把公众号文章、行业分析、产品解读、汽车评论等长文，稳定转换成竖版短视频的 Codex Skill。它不只是做幻灯片，而是把文章逻辑拆成“封面、图解、动画、字幕、配音、BGM、结尾、发布文案”的完整生产流程。
This Skill helps Codex create a repeatable article-to-video workflow:
Article
-> storyboard
-> narration
-> Q-comic images
-> animated explainer
-> synced captions
-> BGM mix
-> upload-grade vertical MP4
The default video structure is:
Full 9:16 Image 2 cover
↓
Body video stack
↓
Full 9:16 Image 2 ending card
The body stack is:
Compact topic bar
Current-article Q-comic infographic
Dark animated explainer board
Bottom narration-synced subtitles
The key idea is simple:
The voiceover, image, middle explainer, and subtitles should explain the same idea at the same time.
That rule prevents the common AI-video problem where the audio says one thing, the image shows another, and the motion graphics are just decoration.
This Skill is useful for:
- WeChat public-account writers repurposing articles into short videos.
- Automotive commentators making daily market or product analysis videos.
- Product analysts explaining launches, pricing, strategy, or user pain points.
- AI content operators building article-to-video pipelines.
- Small teams that need repeatable daily vertical-video production.
- Solo creators who want consistent style without redesigning every episode.
It was tuned through automotive commentary videos, but the structure can be adapted to:
- AI tool tutorials,
- consumer electronics analysis,
- finance or business explainers,
- SaaS product updates,
- company newsletters,
- local service reports,
- educational micro-lessons,
- personal IP commentary.
Long articles often fail when converted to short video because:
- the script is too long,
- the video becomes a boring PPT,
- captions do not match the voice,
- images are nice but do not explain the argument,
- animated panels are generic,
- the cover is just a screenshot,
- the ending feels unfinished,
- every episode has a different visual style.
This Skill turns the process into a stable production system:
- Extract the article's core logic.
- Split it into 5-6 short-video chapters.
- Write a natural voiceover script.
- Generate topic-specific Q-comic infographics.
- Build a dark animated logic board for each chapter.
- Sync captions with the spoken narration.
- Add a standalone cover and ending card.
- Loop and mix BGM under narration.
- Render, inspect, and deliver an upload-grade MP4.
A typical output project may look like this:
assets/
qcovers/
topic-cover-image2.png
qcards/
segment-01.png
segment-02.png
qendings/
topic-ending-image2.png
audio/
narration.wav
bgm.m4a
data/
storyboard.json
audio-sync.json
renders/
topic-final-upload.mp4
qa-cover-frame.png
qa-body-frame.png
qa-ending-frame.png
The exact folder names can change. The production logic matters more than the file layout.
The cover and ending should be complete 9:16 Image 2 images.
Do not make them by:
- cropping horizontal cards,
- stitching screenshots,
- exporting HTML blocks,
- reusing a body frame.
They should work as standalone shareable poster images.
The same video may be posted on WeChat Channels, Douyin, Xiaohongshu, Bilibili, TikTok-style feeds, or other platforms.
Therefore the image itself should not contain platform-specific wording unless the creator explicitly asks for a platform-specific variant.
Avoid text like:
视频号解读抖音专用小红书笔记快手B站TikTok
Use neutral labels instead:
本期解读重点清单风险清单购车前先看一图看懂
This is especially important because image generation models may accidentally add platform labels into corner badges or stickers. Always inspect the generated image before rendering.
The middle board is not a subtitle and not a decorative dashboard.
It should show:
- chapter title,
- one summary sentence,
- 3 nodes or checkpoints,
- quick highlights that follow the narration.
The animation does not need word-perfect sync. It should be reasonably aligned with the voiceover and move fast enough for short-video pacing.
The bottom subtitle strip should show the actual spoken narration clause.
Do not put a separate summary there. The summary belongs in the middle explainer.
If the BGM file is shorter than the video, loop it. A common production bug is that BGM only plays for the first segment and disappears later. This Skill explicitly checks for that.
Clone this repository:
git clone https://github.com/hh19880814-bot/wechat-videohao-qvideo-skill.gitCopy or symlink the Skill into your Codex skills folder:
mkdir -p ~/.codex/skills/wechat-videohao-qvideo
cp SKILL.md ~/.codex/skills/wechat-videohao-qvideo/SKILL.mdRestart Codex or refresh the Skill list if needed.
In Codex, you can say things like:
Use wechat-videohao-qvideo to turn this WeChat article into a vertical short video.
按照 wechat-videohao-qvideo 这个 skill，把这篇公众号文章做成 9:16 视频。
用这个 skill 做一个汽车行业分析视频，封面、正文、结尾都要完整。
继续按这个 Q-video workflow，做今天这篇文章的视频。
If your Codex environment supports Skill metadata, the Skill can also trigger when the task mentions:
- WeChat article to video,
- 公众号文章转视频,
- 视频号竖版视频,
- 抖音短视频,
- Q-comic infographic video,
- dark middle explainer,
- narration-synced captions.
At minimum:
- article text, Markdown, JSON, or a WeChat draft export;
- target topic or title;
- desired publishing platform if it is platform-specific.
Recommended:
- fixed creator voice reference,
- preferred BGM,
- previous accepted video sample,
- existing Q-comic article images,
- desired length.
If you do not have all of these, the Skill can still guide the workflow. It will ask for missing assets instead of silently switching to unrelated placeholders.
Codex extracts:
- what happened,
- why it matters,
- the central contradiction,
- what users should pay attention to,
- the final judgment.
Each chapter gets:
- short title,
- one summary sentence,
- 3 small nodes,
- narration paragraph,
- subtitle clauses,
- image direction,
- middle explainer direction.
Use ChatGPT Image 2 / Image 2 for:
- full vertical cover,
- body Q-comic cards,
- full vertical ending card.
The style should be:
- Q-style hand-drawn infographic,
- stable creator character,
- readable Chinese text,
- visual metaphor,
- article-specific logic,
- no stale labels from old topics.
Use a consistent creator voice when possible.
If using local voice cloning, generate audio first and derive scene timing from actual audio durations. Do not rely only on estimated word count.
Default layout:
topic bar
Q-comic card
dark explainer
caption strip
Keep the stack in the mobile-safe center area. Leave breathing room above and below so platform UI does not cover the core content.
Recommended output:
1080x1920- H.264 video
- AAC audio
- about
10-12 Mbpsvideo bitrate for upload - BGM volume around
0.10-0.12under narration
Before delivery, inspect:
- first frame,
- early body frame,
- middle body frame,
- late body frame,
- ending frame,
- audio track,
- BGM continuity,
- subtitle readability,
- platform-specific unwanted text.
Use wechat-videohao-qvideo.
Article title:
德国公司想卖小米汽车？先看授权和风险
Article summary:
德国公司计划进口中国电动车，小米官方否认合作并启动法律程序。请把重点放在：海外热度、官方授权、非官方进口风险、售后质保、软件法规、本地服务体系。
Requirements:
- 9:16 vertical video
- Image 2 cover and ending
- Q-comic body cards
- dark middle explainer
- real narration subtitles at the bottom
- BGM should loop through the whole video
- do not put platform-specific words in the images
- deliver MP4, short title, description, and tags
A finished video package should include:
- final MP4,
- short title,
- video description,
- topic tags,
- optional standalone cover image,
- confirmation that BGM continues,
- confirmation that cover and ending contain no platform-specific text.
The biggest value of this Skill is not one template. It is the production discipline:
- article logic first,
- voice and visuals aligned,
- cover and ending treated as real assets,
- platform-safe image text,
- audio checked before delivery,
- upload version separated from preview version.
That discipline is what makes AI-assisted short-video production feel less random and more like a repeatable creator workflow.
MIT. Use it, modify it, and adapt it to your own creator workflow.