gemini-video-understanding_skill

This skill analyzes videos with Google's Gemini API to summarize, answer questions, transcribe with visuals, and extract timestamped insights.
  • Python

116

GitHub Stars

6

Bundled Files

2 months ago

Catalog Refreshed

4 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill einverne/dotfiles --skill gemini-video-understanding

  • .env.example108 B
  • EXAMPLES.md7.9 KB
  • QUICKSTART.md2.3 KB
  • README.md3.6 KB
  • requirements.txt20 B
  • SKILL.md10.4 KB

Overview

This skill analyzes videos using Google’s Gemini API to summarize content, answer questions, transcribe audio with visual context, and extract timestamped references. It supports clipping, processing YouTube URLs, multiple video inputs, and a range of Gemini models with large context windows for long videos. The tool accepts nine common video formats and can adapt frame sampling to control cost and fidelity.

How this skill works

The skill accepts local video files or YouTube URLs, uploads large files via the Files API when needed, or sends inline bytes for small files. It runs analysis with a chosen Gemini model, returns structured outputs (summaries, QA, transcriptions with timestamps and visual descriptions), and can export clips or save responses to a file. The script handles API-key detection, polling for file readiness, and provides options for start/end offsets and FPS sampling.

When to use it

  • Summarize a long tutorial, lecture, or meeting with timestamps and key takeaways
  • Transcribe interviews or presentations with speaker notes and visual descriptions
  • Answer specific questions tied to exact moments in a video (MM:SS referencing)
  • Compare multiple product demos or versions of the same footage
  • Create quizzes, study guides, or educational materials from video content

Best practices

  • Use the Files API for videos larger than 20MB to avoid inline limits and reuse uploads
  • Clip long videos to relevant segments to reduce tokens and speed up responses
  • Lower FPS sampling for static content and raise it for action-heavy footage to balance cost and accuracy
  • Choose a 2.5-series model for richer comparisons and longer context; use flash variants for faster results
  • Cache file URIs and reuse them for multiple queries to avoid repeated uploads

Example use cases

  • Generate a 3-point summary with timestamps for a 45-minute tutorial
  • Transcribe a 30-minute interview with speaker labels and visual gesture notes
  • Clip 40s–80s from a demo and summarize only that segment
  • Compare two product demo videos and list differences in feature coverage and clarity
  • Analyze a YouTube lecture URL to produce study questions and answers

FAQ

MP4, MPEG, MOV, AVI, FLV, MPG, WebM, WMV, and 3GPP are supported.

How do I provide the Gemini API key?

Set GEMINI_API_KEY in the environment or place it in a .env file in the skill directory or project root; the script checks these locations in order.

When should I use low-res mode?

Use low-res (fewer tokens per second) to extend context for very long videos or when visual detail is not critical.

Built by
VeilStrat
AI signals for GTM teams
© 2026 VeilStrat. All rights reserved.All systems operational