The Challenge

Our client received hundreds of video submissions per cohort and processed every one manually. Staff would watch each video to write a title, find a usable thumbnail frame, design the thumbnail, and publish. With volume growing each year, publishing delays stretched to days and staff time was increasingly consumed by work that didn't require human judgment.

Specific manual steps involved for each video:

• Watching the full video to write an appropriate title
• Scrubbing through frames to find a visually usable thumbnail moment
• Designing a thumbnail individually for each submission
• Manually entering metadata and publishing

Our Solution

We built a fully automated video processing pipeline that ingests raw video submissions and outputs a publication-ready result — title, thumbnail, and metadata — with no human review needed per video.

• Automatic transcription: Each video is transcribed as it enters the pipeline, producing a clean text record of the content.
• AI-generated titles: The transcription is analyzed to generate a relevant, accurate title without anyone needing to watch the video.
• Intelligent frame selection: The system scans the video to identify the most visually suitable frame for a thumbnail based on image quality and composition.
• Templated thumbnail generation: The selected frame is formatted using one of 7–8 pre-designed visual templates, ensuring consistent, professional-looking thumbnails across all published videos.

Technologies Used

• Whisper (speech-to-text transcription)
• Python (video processing pipeline)
• YouTube APIs
• Frame extraction
• Thumbnail template system

Results

• Publishing time reduced from days to minutes per video
• Scalable to hundreds of videos per day without adding staff
• 7–8 visual templates ensure consistent branding across all published content

‍