How-to guides

Task-oriented recipes. Each one assumes you already know the basics from the tutorials and gets straight to the goal.

Editing

• Build styled excerpts and summaries — selected footage with captions, a logo and title, or a ducked music bed.
• Process hour-long videos — bounded frame buffers, frame iteration, and the memory-efficient dubbing path.

AI

• Assemble a video from AI-generated media — text to image to video, with narration and crossfades.
• Dub a video into another language — translate, clone the voice, and re-time it onto the source.
• Update dubbing timing consumers — migrate timing calls and saved results to 0.61.2.

Subtitling is covered end to end in Tutorial 2.

Putting an LLM in the loop

There are three ways, and they differ in who owns the model:

You want…	Use
videopython to edit for you, fully local	Let a local LLM edit for you
your own agent/harness to drive the tools	Drive editing from an MCP agent
to author and validate plans from your own LLM integration	Author edit plans with your own LLM

To inspect and replay a complete JSON-plan example without a model, see Reproduce the agent-authored edit demo.

The design behind all three is described in LLM-first design.

Verification

• Verify local AI models — run the repository harnesses and record model checks.