--video-mode transcript prefers transcript-first media handling even when a page has text.
Direct media URLs (mp4/webm/m4a/etc) skip HTML and transcribe.
Local audio/video files are routed through the same transcript-first pipeline.
--diarize [auto|elevenlabs|openai] adds speaker labels to local files and direct media URLs; --identify-speakers, anchors, profiles, and remembered mappings use the same naming pass as YouTube. Local files do not require yt-dlp. Local video is converted once to mono 16 kHz MP3 before upload, then reused if diarization falls back between providers.
YouTube diarization selects audio-only media. Combined --slides --diarize downloads separate video-only and audio-only streams in one yt-dlp run, then shares the video with slide extraction.
YouTube still uses the YouTube transcript pipeline (captions → yt-dlp → Android VR direct audio fallback).
X/Twitter status URLs with detected video auto-switch to transcript-first (yt-dlp), even in auto mode.
X broadcasts (/i/broadcasts/...) are treated as media-only and go transcript-first by default.
Loom share/embed recording URLs use the generic yt-dlp + transcription path. Explicit --video-mode transcript requests never fall back to Loom landing-page text; guarded daemon flows enable the external downloader only for strict Loom recording URLs in this explicit mode.
Local media files are capped at 2 GB. Remote podcast/media transcription downloads are capped at 512 MB by default and fail closed with Remote media too large even when the server omits or under-reports Content-Length; other remote media URLs are best-effort via yt-dlp.
Operators who accept the disk/DoS tradeoff for larger remote podcast/media files can opt in with SUMMARIZE_REMOTE_MEDIA_MAX_BYTES=<positive integer byte limit>. The override must be a finite positive integer byte count; fractional, sub-byte, or otherwise invalid values are ignored and the default 512 MB cap remains in effect.
Remote transcription providers: ASSEMBLYAI_API_KEY, GEMINI_API_KEY / GOOGLE_GENERATIVE_AI_API_KEY / GOOGLE_API_KEY, OPENAI_API_KEY, FAL_KEY, DEEPGRAM_API_KEY (plus GROQ_API_KEY before local/remote fallback). Gemini defaults to gemini-2.5-flash; override with SUMMARIZE_GEMINI_TRANSCRIPTION_MODEL. Deepgram defaults to nova-3; override with SUMMARIZE_DEEPGRAM_TRANSCRIPTION_MODEL. Diarization uses ELEVENLABS_API_KEY or OPENAI_API_KEY.
Gemini uses the Files API automatically for larger uploads.
When media is detected on a page, the Summarize button gains a dropdown caret (Page/Video or Page/Audio).
Selecting Video/Audio forces URL mode + transcript-first extraction for that run only.
Selection is not stored.
Chrome Browser mode transcribes fetchable direct and embedded media in bounded MediaBunny/WebCodecs chunks with browser-cached multilingual Whisper Tiny. YouTube prefers active-player/watch-page direct audio, then Android VR, buffered direct audio, and captured SABR.
Browser slide extraction uses ranged MediaBunny URL reads instead of buffering the complete video. The Whisper runtime is disposed after an idle period while the downloaded model remains in browser cache.