Descriptions:
Web Dev Cody continues his ongoing series on building an AI-powered video generation SaaS using agentic coding tools. Part 2 focuses on two concrete engineering tasks: migrating the codebase to TypeScript and fixing a critical timing bug where generated videos were cutting off early because insufficient video segments were being created to cover the full audio duration.
The video walks through diagnosing a mismatch between a 23-second MP3 output and a 16-second final video, then prompting the AI coding agent to implement an OpenAI Whisper-based solution. The fix uses Whisper’s timestamp data to dynamically calculate exactly how many 5-second WAN 2.2 video clips are needed to cover the full script, with a freeze-last-frame fallback for any remaining audio tail. Additional features scoped out during the session include ElevenLabs voice actor selection with preset IDs, FFmpeg-based burned-in captions, and direct YouTube publishing as future roadmap items.
The episode offers a realistic look at the iterative agentic coding workflow — writing user stories, issuing natural-language prompts to Claude Code or Cursor, reviewing generated plans, and babysitting execution. Developers building multi-API AI pipelines or learning to structure agentic coding projects around FAL AI, OpenAI, and ElevenLabs will find this series a practical reference.
📺 Source: Web Dev Cody · Published February 17, 2026
🏷️ Format: Hands On Build
