Cole Medin

This Skill Turns Claude Code Into an Expert Video Editor (I'm Giving It Away)

Published 2026-11-19 · Duration 19:20

Summary

This presentation details an advanced AI-driven video editing pipeline, where a 'skill' guides a coding agent (like Claude Code) to transform raw footage into polished content. The core process involves a scripted sequence of steps—transcription, cutting, assembly, and polishing—which can be augmented using multimodal APIs (OpenRouter) for complex tasks like voice cloning and thumbnail generation.

Download summary

Key takeaways

  1. The Core Editing Pipeline 1:40

    The process is scripted and deterministic, moving from raw input to final export through five main stages: transcription (with timestamps), retake removal/cutting (dead air, breaths), assembly, polishing (transitions/animations), and final validation.

  2. Multimodal Augmentation 10:30

    The 'full pipeline' utilizes OpenRouter to access multimodal capabilities that local LLMs lack, including video understanding, image generation (for thumbnails), and voice cloning (e.g., Fish Audio 2.1 Pro).

  3. Customization and Flexibility 6:10

    The skill is designed to be modular; users can customize the process by adjusting the steps, adding new steps, or modifying 'lessons.mmd' (hard rules) and 'brand.json' (colors) to fit specific content types (e.g., talking head, screen+face panel, shorts).

  4. AI-Detected Fixes

    The system demonstrated the ability to detect and fix subtle issues, such as small stumbles mid-word or choppy audio, through self-iteration and validation steps.

More digests on AI

Technical details

  • Skill Architecture 270s

    The `edit-video` skill guides the coding agent through a sequence of helper scripts. These scripts handle deterministic tasks (e.g., transcription, pause detection) that an LLM cannot perform alone, providing data back to the main orchestrator.

  • Validation Process 450s

    A validation step involves retranscribing the output and performing a word-by-word diff against the source transcripts to ensure cuts were made at the correct timestamps.

  • Model Usage 670s

    The full pipeline utilizes specific models via OpenRouter for complex tasks: Gemini 4 Argon for video review, Fable 5.2 for planning, and Fish Audio 2.1 Pro for voice cloning.

Mentioned resources

Channel & topics

Watch on YouTube · Back to latest

This independent, AI-assisted summary is provided for commentary and informational purposes. It may contain errors or omit important context. Please watch the original video for the creator's complete presentation. Video, thumbnail, and related copyrights belong to their respective owners.