YouTube Video to Blog Post: Gemini AI Transcription

YouTube link → download audio → Gemini transcribe → AI generates structured blog → output JSON

Who

Content creators, marketers, educators, bloggers

You get

Automatically convert any YouTube video into a blog post with title, description, category, tags, and full transcript

You need

n8n · yt-dlp

Steps

1. Configure webhook trigger to receive YouTube URL

   Set up a Webhook node in n8n with method POST, path 57fddbda-a118-4546-a223-8e91a7d53ff6, and responseMode set to responseNode. The caller sends a JSON body: {"videoUrl": "https://www.youtube.com/watch?v=..."}. The workflow starts from this node.

2. Generate timestamp

   Use a Set node to generate the current Unix timestamp (now_unix = {{ new Date().getTime() }}) for unique audio file naming in subsequent steps.

3. Download YouTube audio

   Use an Execute Command node to run yt-dlp, extracting audio to MP3. The command references the upstream videoUrl and timestamp. Ensure yt-dlp is installed on the n8n host.

   Execution prompt · Source text。Execute Command node Command field

   Copy prompt

   =yt-dlp -x --audio-format mp3 -o output-{{ $json.now_unix }}.mp3 '{{ $('Get Youtube Url').item.json.body.videoUrl }}'

4. Read MP3 binary file

   Use a Read/Write Files from Disk node to read the generated MP3 file (file selector: =output-{{ $('Get Time').item.json.now_unix }}.mp3) as binary data, passing it to the transcription node.

5. Transcribe audio with Gemini

   Use a Google Gemini node (models/gemini-1.5-flash) with binary audio input to transcribe speech to text. Requires googlePalmApi credentials, resource set to audio, inputType set to binary.

6. AI Agent generates structured blog post

   The AI Agent node receives the transcript and generates a blog post via Google Gemini Chat Model with Structured Output Parser enforcing a fixed JSON Schema. The Structured Output Parser has autoFix enabled and provides jsonSchemaExample as a format reference. Includes both user prompt and system prompt.

   Execution prompt · Source text。AI Agent node Text / User Message field

   Copy prompt

   =Please create a blog post from the following video transcript. Identify the key themes, structure the content logically, and generate a title, category, and tags.
   
   Transcript:
   {{ $json.content.parts[0].text }}

   Execution prompt · Source text。AI Agent node Options > System Message field

   Copy prompt

   =You are an expert content strategist and editor. Your task is to transform a raw video transcript into a well-structured and engaging blog post.
   
   You must adhere to the following rules:
   1.  Your entire response must be a single JSON object enclosed in a markdown code block (```json ... ```).
   2.  Do NOT include any introductory phrases, explanations, apologies, or any text whatsoever outside of the JSON markdown block.
   3.  The tone should be informative, clear, and engaging, transforming the conversational transcript into polished written content.
   4.  The language of all generated text in the output JSON must match the primary language of the input transcript.
   5.  The JSON object must have this exact structure:
   {
     "category": "A single, top-level category based on the content.",
     "tags": [
       "An array of 3 to 5 relevant, lowercase string tags."
     ],
     "title": "A creative, compelling title for the blog post.",
     "short_description": "A concise, one-to-two sentence summary of the blog post, optimized for search engines (around 150-160 characters).",
     "body": "The full blog post content formatted in HTML, including <h1>, <h2>, and <p> tags."
   }

7. Merge video URL and full transcript

   Use a Set node to append the original YouTube URL (videoUrl = {{ $('Get Youtube Url').item.json.body.url }}) and full transcript text (fulltext = {{ $('Transcribe a recording').item.json.content.parts[0].text }}) to the AI Agent output, enriching the final payload. includeOtherFields is set to true to preserve upstream fields.

8. Return blog post JSON

   Use a Respond to Webhook node to return the final structured JSON (blog post, category, tags, transcript, original video link) to the caller, or forward it to a configured CMS/database endpoint.

You are an n8n workflow setup assistant. The user wants to replicate the "YouTube video to blog post" template. First confirm: 1) whether the user has shell access to their n8n instance to install yt-dlp; 2) whether they have a Google Gemini API key; 3) their target output endpoint (CMS, Notion, Slack, or other webhook). Then guide them to import the workflow JSON or build nodes manually, replacing credentials and the output endpoint.