[ CONTENT_PRODUCTION ]

Frame-by-Frame Video Analysis Skill

2026Developer PlatformsCapability

What it is

Any video URL turned into a timestamped structural breakdown.

Client work is shown without identifying imagery

[ THE_PROBLEM ]

Why this existed

Understanding why a piece of video works means watching it repeatedly and taking notes, which nobody does systematically enough to learn from.

[ WHAT_WE_BUILT ]

What we built

A packaged agent skill that downloads a video from YouTube, Instagram or any supported site, uploads it to a multimodal model, polls until processing completes, and streams back a per-interval breakdown: timestamp, what is happening, scene description covering setting, lighting, colour, composition and mood, and an image-generation prompt to recreate that exact frame. Self-bootstrapping — the setup script creates the environment, installs dependencies and scaffolds config on first run.

Downloads video from YouTube, Instagram or any supported site
Uploads to a multimodal model and polls until processing completes
Emits per-interval breakdown: timestamp, action, scene description, image prompt
Scene descriptions covering setting, lighting, colour, composition and mood
Self-bootstrapping setup creating the environment and scaffolding config on first run

[ HOW_IT_IS_USED ]

How a company uses it

A content team reverse-engineers what makes a competitor's video work — shot by shot — instead of watching it four times and taking notes.

Built with

PythonGeminiyt-dlpffmpeg

[ COMMON_QUESTIONS ]

Questions clients ask

What is the image-generation prompt for?

It turns analysis into production input. Having a prompt that recreates a specific frame means the breakdown feeds directly into storyboarding or reference generation rather than sitting in a document.

Why package it as an agent skill?

So it is invoked in the flow of work — asking for a breakdown in the tool you are already using — rather than being another application to remember exists. Self-bootstrapping matters for the same reason: a setup step is where adoption dies.

Can it analyse long videos?

Yes, though processing time and cost scale with length. For competitive analysis the useful unit is usually a short-form video or a segment rather than a full episode.

[ RELATED_WORK ]