How the Video Summarizer Works
AnyTool summarizes video entirely in the browser. It accepts a video/audio file, an SRT/VTT upload, or a pasted transcript; for media it decodes the audio with the Web Audio API and transcribes it with a self-hosted Whisper model (WebGPU with WASM fallback) to get timestamped text. A hybrid pipeline windows the transcript, runs an extractive pre-filter (TF-IDF/TextRank), and feeds a self-hosted abstractive summarization model (distilbart/T5 via transformers.js) in a map-reduce to produce per-chapter blurbs and a global TL;DR; topic-shift detection between windows yields clickable timestamped chapters that seek the in-page player. Outputs export to Markdown, TXT and SRT/VTT.
- Runs client-side — video and transcript are never uploaded
- Self-hosted Whisper transcription — works without captions
- Abstractive TL;DR + key points via map-reduce
- Clickable timestamped chapters (seek the player)
- Inputs: video/audio, SRT/VTT, pasted transcript
- No upload, no caps, no sign-up
