Douyin Copy Ingestion from Feishu

Problem

When Douyin video clips are already stored as attachments in Feishu Bitable, manually opening each video, transcribing the speech, and reformatting category or viral-copy analysis becomes repetitive and easy to standardize poorly.

How It Works

This skill treats a Feishu Bitable attachment field as the input source. It fetches temporary file download URLs through the Feishu API, saves the videos to a local temporary directory, transcribes audio with SenseVoiceSmall, segments the transcript with Qwen2.5-7B-Instruct, infers the Douyin three-level category from the title and content, analyzes the copy across nine viral-logic dimensions, and writes the results into the target Bitable fields such as category, raw transcript, segmented copy, and analysis. In batch mode, it iterates over source-table records, skips videos whose files already exist locally, and pauses for one second between items to reduce rate-limit pressure.

Boundaries

It fits workflows where the video files are already in a Feishu table, not the single-URL Douyin link workflow. It requires Feishu app credentials, writable target-table fields, and access to the speech and model APIs used for transcription and analysis. When the source and destination tables differ, the source-table token and ID must be configured separately from the target-table values.