watch

name	watch
description	Analyze videos, screen recordings, ads, and video links with LimeBot's native video tool. Use for requests to watch, summarize, transcribe, inspect a moment, read on-screen text, or answer questions about YouTube, Vimeo, TikTok, Loom, MP4, MOV, MKV, or WebM media.
dependencies	{"python":["yt-dlp","Pillow"],"binaries":["ffmpeg","ffprobe"]}
metadata	{"aliases":["video"],"keywords":["watch","video","transcript","screen-recording"]}

Watch videos

Call analyze_video; never shell out or call yt-dlp/FFmpeg directly.

• Use detail="transcript" when visuals are irrelevant.
• Use detail="efficient" for a quick visual scan.
• Use detail="balanced" by default for summaries, UI recordings, ads, and visual questions.
• Pass start and/or end whenever the user names a moment or time range.
• Use resolution=1024 only when small on-screen text must be read; otherwise use 512.
• Set question to the user's actual question so long transcripts retain relevant segments.
• Treat contact sheets as internal visual context; do not send them unless the user asks.

When answering, distinguish evidence from native captions, OpenAI Whisper, and frames-only analysis. If the tool reports missing dependencies, relay its exact installation command. If captions are absent and Whisper is disabled, do not imply that audio was understood.