Douyin Video Storyboard Extraction to Feishu

Problem It Solves

Douyin short-video analysis often stops at “this feels good, but why?” Lines, pacing, visual emphasis, and conversion timing are scattered across the video, and manual spreadsheets easily miss opening hooks, subtitle boundaries, and shot changes.

How It Works

The skill turns a video into an auditable storyboard table and writes the result into a Feishu Bitable.
- Frame extraction: samples 0-5s at 1 fps, captures body frames by subtitle sentence, and merges frames closer than 0.8s.
- Semantic segmentation: aligns timeline, dialogue, and visual content using speech-to-text, segmentation, and frame description.
- Viral-logic analysis: evaluates opening hooks, visual rhythm, emotional curve, and conversion timing for each storyboard row.
- Feishu write: batch-writes fields such as Video ID, Video Name, Start Time (s), End Time (s), Dialogue, Visual Description, and Viral Role, then uploads screenshot attachments.

Boundaries

It requires a SiliconFlow API Key and Feishu credentials; the target Bitable must exist and be configured with App Token and Table ID. Visual-description quality depends on the supported multimodal model, so shot details may remain coarse if the model only outputs basic descriptions. It fits teardown, review, and asset archiving, not editing, media buying strategy, or brand copywriting.