Creator Use Cases In development
Three real-world pipeline cases, spanning video, long-form articles, and short social content.
Case 1
Bilibili Video Commentary
A single Bilibili video link → voiceover script + cover copy.
| Stage | Tool | Output |
|---|---|---|
| Input | Paste video link | Original video URL |
| Download | yt-dlp | mp4 file |
| Transcribe | Whisper large-v3 | Chinese subtitles with timestamps |
| Refine | LLM (DeepSeek-Chat) | Segmented summary + highlights |
| Rewrite | LLM (Claude Sonnet) | 3 voiceover scripts (serious / playful / suspenseful) |
| Cover | LLM | 3 candidate cover titles |
| Output | Creator editor | Ready to paste into the Bilibili dashboard |
Typical Time
10-minute video → full pipeline in 90 seconds.
Case 2
Long-Form WeChat Article
A single topic → a 2,000-word in-depth WeChat article.
graph TD
TOPIC["Topic input: A retrospective on applying WAL to AI Agent persistence"] --> SEARCH[Source retrieval<br/>Lurus API + web_search Tool]
SEARCH --> OUTLINE[Outline generation<br/>LLM drafts 5 H2 headings]
OUTLINE --> WRITE[Section-by-section writing<br/>Generated in chunks per the outline]
WRITE --> FIG[Figure suggestions<br/>LLM proposes 3 diagram placements]
FIG --> MD[Markdown output to the Creator editor]Adaptations:
- WeChat does not support H4, so it is automatically downgraded
- Automatically generates "recommended reading" cross-links (based on keyword matching)
- Supports switching between three style tiers: "conversational / academic / commercial"
Case 3
Xiaohongshu Note
One image + one quip → a complete recommendation note.
Input: Product photo + "the battery life is unreal"
Output:
┌─────────────────────────────┐
│ 🌙 终于找到通勤救星！ │
│ │
│ 用了一周真的爱上了...（100 字）│
│ │
│ ✅ 续航 28 小时 │
│ ✅ 重量只有 180 克 │
│ ✅ 降噪 -35dB │
│ │
│ #数码好物 #通勤 #降噪耳机 │
└─────────────────────────────┘Automatically generates 6–10 hashtags while avoiding banned words.
Shared Pipeline Comparison
| Scenario | Video Commentary | Long-Form WeChat Article | Xiaohongshu |
|---|---|---|---|
| Input form | URL | Topic keyword | Image + short caption |
| Core model | Whisper + LLM | LLM | LLM |
| Output length | 300–500-word script | 2,000 words | 80–150 words |
| Publishing support | Copy & export | Auto-publish via chromedp | Auto-publish via chromedp |
| Typical time | 90 seconds | 2 minutes | 30 seconds |