Introduction¶

Content creators and asset managers often encounter a specific need: they have a video from Douyin, Bilibili, or YouTube, and want to convert the audio into searchable, editable text. The manual workflow typically involves downloading the video, launching a transcription tool, and saving the text separately—a fragmented process with a long pipeline. What’s more, videos often contain subtitles and background music in addition to human voices, leading to homophone errors in raw ASR (Automatic Speech Recognition) output.

If this workflow could be triggered by a single command to an Agent—inputting a link and receiving text immediately—the value would be significant. The “video-to-transcript” introduced below is a Skill designed for exactly this purpose: it receives video links within an Agent conversation, transcribes the audio, and outputs both a proofread plain text version and a timestamped text version.

What is this¶

video-to-transcript is a Skill on the SkillHub community platform, identified by the coordinate @user_ea76c6e6/video-to-transcript. It is published by user_ea76c6e6, is currently at version 1.3.2, and falls under the content-creation category. It solves the following problem: when a user says something like “transcribe this video” in a conversation, the Agent automatically handles link retrieval, speech transcription, and (optionally) homophone proofreading, returning usable text.

According to data on the SkillHub page, this Skill has a rating of 100,000 and 2,005 downloads. The page tag is requires_api_key=false, meaning it can run without configuring an API key.

Core Features¶

The following capabilities have been verified from the official description:

1. Link Transcription: Simply input a video link. The official description lists support for Douyin/Bilibili/YouTube.
2. Transcription Engine: Uses faster-whisper for transcription.
3. LLM Homophone Proofreading (Optional): Enabled when an API key is configured; outputs raw ASR text when not configured.
4. Two Output Modes: Proofread plain text and timestamped text.
5. Local Execution: Except for LLM proofreading, transcription is completed locally, with the official description emphasizing privacy and security.

Trigger Words: The trigger scenarios listed in the Skill description include “transcribe this video,” “video to text,” “Douyin transcription,” “extract video text,” “generate subtitles,” and “turn video into text.” Note that this description does not perfectly match the summary captured from the page (the page summary was truncated), so it is recommended to verify on the official page before integration.

The same Skill description also mentions support for Xiaohongshu (RED), TikTok, and local video/audio files, claiming parallel processing of transcription and OCR extraction of hard subtitles. These points did not appear in the captured page summary and are considered items to be verified; they will be mentioned again in the “Use Cases” section below.

Installation and Activation¶

This Skill is a SkillHub community Skill; the installation entry point is subject to the official page:

https://www.skillhub.cn/skills/user_ea76c6e6/video-to-transcript

Coordinates and Classification (for search or integration):

@user_ea76c6e6/video-to-transcript
content-creation
requires_api_key=false

Note: The captured data does not provide specific CLI installation commands or directory paths, so none are fabricated here. Once activated, the Skill will be invoked when the aforementioned trigger words appear in the Agent conversation.

Typical Usage¶

Directly express your need in the Agent conversation, for example:

Transcribe this video: https://www.bilibili.com/video/BVxxxxxx

Upon receiving the trigger word, the Agent calls this Skill to output proofread plain text. If a timeline is needed, you can request the timestamped text. LLM proofreading is an optional step: if no API key is configured, the output will be the raw ASR text from faster-whisper.

Use Cases and Notes¶

Suitable for: Content creators organizing short video assets, converting videos to searchable text, preparing drafts for subsequent summarization or secondary creation; as well as developers looking to add a “video in, text out” preliminary step to their Agent workflows.

Notes:

1. Recognition quality depends on the audio itself. In scenarios with overlapping voices, loud background music, or heavy accents, ASR errors will be more frequent. This is exactly what the LLM homophone proofreading aims to solve, but it is an optional capability.
2. Proofreading relies on an LLM API key; without configuration, only raw ASR text is available.
3. Transcription is completed locally, making it suitable for privacy-sensitive scenarios, but the local environment must be capable of running faster-whisper and its dependencies.
4. Support for Xiaohongshu, TikTok, local files, and parallel OCR hard subtitle extraction comes from the Skill description rather than the captured page summary. It is recommended to confirm these on the official page before integration.

Summary¶

video-to-transcript turns “give a link, get text” into a single step within an Agent: local transcription via faster-whisper, optional LLM proofreading, outputting plain text and timestamped text, all running locally. Developers who need to build their own transcription workflows can use this as a starting point.

Official Address:

https://api.skillhub.cn/user_ea76c6e6/video-to-transcript

SkillHub Page:

https://www.skillhub.cn/skills/user_ea76c6e6/video-to-transcript