gemini-video-understanding

Gemini Video Understanding analyzes videos with the Gemini API to produce summaries, transcripts with timestamps, and visual context. Configure your GEMINI_API_KEY, then run analyze_video.py with a local file or YouTube URL to get structured analysis outputs including text responses and usage metrics.

Yes, this Skill supports both YouTube URLs and local video files via the Gemini Files API or inline data. You can also compare multiple videos (up to 10 with Gemini 2.5+) and apply clipping with custom start/end offsets and frame-rate sampling.

The Skill handles local video files and YouTube content with configurable frame-per-second sampling, video clipping offsets, and token-context windows up to 2M. It supports Gemini 2.5 and 2.0 model variants for processing.

Submit a question along with your video file or YouTube URL to the analyze_video.py script. Gemini returns precise timestamps and answers grounded in video content, with structured outputs including text responses and token usage.

Set the GEMINI_API_KEY environment variable (via environment, skill directory, or project root), select a Gemini model variant (2.5 or 2.0), and prepare your video source as a local file path or YouTube URL to begin analysis.

Yes, with Gemini 2.5+, you can analyze up to 10 videos simultaneously. The Skill outputs summaries, transcripts, and insights for each video, enabling cross-video comparison and analysis.