How the Text-Based Video Trimmer Works
AnyTool edits video by its transcript entirely in the browser. The video's audio is decoded with the Web Audio API, resampled to 16 kHz mono, and transcribed by a self-hosted Whisper model (transformers.js, WebGPU with WASM fallback) to produce word-level timestamps. Deleting words or sentences in the transcript marks those time ranges as removed; one-click passes remove filler words (regex) and silences (gaps over a threshold). The kept ranges are merged and cut into a real MP4 using the self-hosted ffmpeg engine's trim+concat with re-encode for frame-accurate boundaries; the edited transcript can also be exported as SRT/TXT.
- Runs client-side — transcription and cutting are local
- Self-hosted Whisper (WebGPU/WASM) for word timestamps
- Delete text to cut; remove fillers and silences
- Real MP4 export plus SRT/TXT
- Live preview that skips removed ranges
- No upload, no watermark, no sign-up
