Run the Pipeline

Time to generate your first video. The pipeline runs in stages — you can trigger each one manually to watch content progress.

Before You Start

Make sure you have:

• At least one channel created with sources configured
• Your OPENAI_API_KEY set in the worker environment
• The worker service running

Running Each Stage

Go to the Pipeline page in the sidebar. You’ll see buttons for each stage.

Step 1: Fetch Topics

Click Fetch Topics. This:

• Pulls headlines from all your configured sources
• Filters them through GPT-4o-mini
• Creates draft content items

Check the Library page — you should see new items with status “draft.”

Step 2: Generate Scripts

Click Generate Scripts. This:

• Takes draft items and generates hooks + full scripts
• Sets duration based on word count and speaking pace
• Moves items to scripted status

Go to any content item in the Library to read the generated hook and script.

Step 3: Generate Voice

Click Generate Voice. This:

• Converts each script to spoken audio via OpenAI TTS
• Generates word-level subtitle timing via Whisper
• Uploads the MP3 to storage
• Moves items to voiced status

You can play the audio on the content detail page.

Step 4: Assemble Video

Click Assemble Video. This:

• Composes the final video with FFmpeg (background + audio + subtitles)
• Generates a branded thumbnail
• Uploads the MP4 and thumbnail to storage
• Moves items to review status

Note: Video assembly is memory-intensive. It processes 1 item at a time by default.

Checking Progress

• Library (sidebar) — Filter by status to see items at each stage
• Pipeline page — Shows recent pipeline run history with success/error counts
• Content detail — Click any item to see its current state, play audio/video, read the script

If Something Fails

If a content item fails at any stage:

1. Go to the content detail page
2. The error message will tell you what went wrong
3. Click Retry for the failed stage
4. The item resets to the previous status and re-runs

Common issues:

• Script generation fails: Usually an OpenAI API error — retry typically works
• Voice generation fails: Check that ffprobe is installed (needed for duration detection)
• Video assembly fails: Usually a memory issue — try reducing slideshow image count or disabling slideshow