media-processing

To extract keyframes and output structured JSON, the skill ingests video files via absolute paths, uses ffmpeg-based preprocessing to extract segments and keyframes, and maps them through Gemini structured generation to produce schema-guaranteed JSON outputs.

Analyzing video content for LLM Q&A without reprocessing involves mapping video segments into structured JSON via Gemini, then using Claude to reduce results and answer reusable follow-up queries over the prior map outputs.

Yes, you can use ffmpeg to generate video clips. The skill extracts shareable clips around specified time ranges and saves them as attachments, enabling direct in-chat delivery of the generated video segments.

Yes, multimodal processing supports optional audio transcription alongside video keyframe extraction. During the configurable preprocessing pipeline phase, the skill can ingest audio and transcribe it while simultaneously extracting video segments.

To ingest media files while deduplicating content, the skill registers files from an absolute path, detects MIME type, extracts duration, and automatically deduplicates video and audio assets using a computed content hash.

You should not use Claude for querying over map outputs if your task requires bypassing the Gemini structured generation phase, as the Claude-based querying mechanism depends on prior map-output data to perform natural-language analysis.