# ClipQuill > Free, private video and audio to text. Runs in your browser — the file is processed on your own > device and zero bytes are uploaded. No signup, no account, no daily cap. ClipQuill is a privacy-friendly, browser-based video transcription tool: it converts videos and audio into text locally in your browser, and the file is never uploaded. ClipQuill transcribes a video or audio file you already have. A first visit downloads about 78.4 MiB (82.2 MB) with brotli in transit, and about 97.5 MiB (102.3 MB) once decompressed: 76.0 MiB of model files plus the 21.5 MiB runtime that runs them. Of that, 73.3324 MiB is the model weights themselves, which are already quantised and barely compress. Everything after the first visit happens locally. ## Pages Grouped by what they are for. Every URL below is in https://clipquill.com/sitemap.xml. ### Core tool and differentiation - https://clipquill.com/ : the transcriber itself — drop a file, pick the spoken language, get TXT / SRT / VTT. Runs in the tab, zero upload. - https://clipquill.com/privacy-policy/ : what is and is not sent anywhere — nothing is uploaded; only the model is downloaded. - https://clipquill.com/video-to-text-accuracy/ : the measured accuracy, by acoustic condition, with the clips that went badly included. - https://clipquill.com/transcription-benchmark/ : the benchmark itself — what was run, on what machine, and how to reproduce it. ### Proof and method - https://clipquill.com/video-to-text-test-conditions/ : the conditions each number was measured under. - https://clipquill.com/video-to-text-character-variants/ : Simplified vs Traditional Chinese output, measured on one clip. - https://clipquill.com/video-to-text-questions/ : FAQ. - https://clipquill.com/ai/ : notes on the models and how the site is meant to be read by agents. - https://clipquill.com/about/ : why recognition runs in the browser, and why the numbers ship with their conditions. - https://clipquill.com/contact/ : contact. ### Input formats - https://clipquill.com/mp3-to-text/ : MP3 to text - https://clipquill.com/mp4-to-text/ : MP4 to text - https://clipquill.com/m4a-to-text/ : M4A to text - https://clipquill.com/wav-to-text/ : WAV to text - https://clipquill.com/audio-file-to-text/ : audio file to text - https://clipquill.com/audio-to-text/ : audio to text - https://clipquill.com/audio-to-srt/ : audio to SRT, mp3 to SRT and mp4 to SRT — the 48 kHz WAV master and a FLAC made from it produce the same 895-byte SRT while the two files sit 3.9 times apart, and four MP4s that differ only in the captions already inside them produce one identical 902-byte SRT - https://clipquill.com/mkv-to-srt/ : MKV to SRT — four MKV builds span 7.064 times the bytes and the subtitle file inside every one of them is the same 852 bytes - https://clipquill.com/mp3-to-srt/ : MP3 to SRT — 64, 128 and 320 kbit/s exports of one 52.469-second recording return the same 1,390-byte SRT with SHA-256 bc6824a4 while the audio files are 5.0 times apart, the 32 kbit/s file gets 119 of 133 reference words right against 118, and the 8 kHz file drops to 113 - https://clipquill.com/transcript-editor/ : audio transcript editor and transcript editing software — editing the words never touches the audio, so every cue keeps its recogniser timing: deleting 12 of 48 words left a 2,530 ms hole, the same deletion with later words pulled up moved the reading speed from 13.17-14.04 to 1.69-15.42 characters per second, and replacing one word with five pushed a line to 44 display units; 0 of 42 readable ranking pages prints a cue count or a reading speed ### Platforms and sources - https://clipquill.com/youtube-video-transcript/ : YouTube video transcript - https://clipquill.com/youtube-video-transcript-generator/ : YouTube transcript generator - https://clipquill.com/transcribe-youtube-video/ : transcribe a YouTube video - https://clipquill.com/transcribe-youtube-video-to-text/ : YouTube video to text - https://clipquill.com/how-to-transcribe-a-youtube-video/ : how to transcribe a YouTube video - https://clipquill.com/google-meet-transcription/ : Google Meet transcription - https://clipquill.com/timestamped-transcript/ : timestamped transcript, SRT and VTT export — the line-width budget, the 0.5 s cue floor and the gap between cues, measured on our own output - https://clipquill.com/interview-transcription/ : interview transcription - https://clipquill.com/podcast-transcription/ : podcast transcription - https://clipquill.com/voicemail-to-text/ : voicemail to text - https://clipquill.com/voice-to-text/ : voice to text - https://clipquill.com/tiktok-transcript/ : TikTok transcript — 37 words per 15 seconds, and the music track costs one of them - https://clipquill.com/instagram-transcript/ : Instagram transcript and Instagram reel transcript — seven words written on the Reel, none in the text - https://clipquill.com/instagram-reel-transcript/ : Instagram reel transcript — the file you upload is not the Reel you watched - https://clipquill.com/instagram-duet-transcript/ : Instagram duet transcript — the second voice costs 4.1% taking turns and 44.9% arriving together - https://clipquill.com/instagram-duet-srt/ : Instagram duet SRT — the words went from 3.1% wrong to 73.5% wrong and the subtitle file did not move - https://clipquill.com/instagram-remix-transcript/ : Instagram remix transcript — the music bed under the second voice turns 44.9% wrong into 73.5% wrong - https://clipquill.com/instagram-long-reel-transcript/ : Instagram long Reel transcript — four crossings of one seam cost zero words, and the caption clock runs 14.984 seconds ahead - https://clipquill.com/instagram-srt-vtt/ : Instagram SRT vs VTT transcript — the two files differ by 3n-17 bytes for 10 to 99 cues, and the audio is not in the formula - https://clipquill.com/facebook-video-transcript/ : Facebook video transcript — a music bed at 50% of the audio turns 22.06% wrong into 47.06% wrong - https://clipquill.com/zoom-transcription/ : Zoom transcription — the platform processes the audio before anything reads it - https://clipquill.com/teams-transcription/ : Teams transcription — the same 72.5 seconds of speech, seven ways into a meeting, from 9.5% to 85.7% wrong - https://clipquill.com/call-transcription/ : call transcription — five minutes of hold music costs no words on 189, and the three-times-longer file takes 86.2 seconds instead of 53.4 - https://clipquill.com/deposition-transcription/ : deposition transcription — 99% on a 189-word passage allows 1.89 wrong words, and of the 16 pages we could read 0 print the word count behind the percentage they quote - https://clipquill.com/voice-recorder-to-text/ : voice recorder to text — seven ways to store the same 72.5 seconds, and only the noise reduction button costs words - https://clipquill.com/caption-generator/ : subtitle maker and caption generator — reading speed runs 15.38 to 19.10 characters a second on the same 52.049 seconds of speech - https://clipquill.com/spotify-transcript/ : Spotify transcript — a chapter mark that misses the pause drops 4 of 189 words, and the subtitle file reads the same speed either way ### Keyword entry pages - https://clipquill.com/transcribe-video-to-text/ : transcribe video to text - https://clipquill.com/transcribe-video-to-text-free/ : free video to text, no signup - https://clipquill.com/video-to-text/ : video to text - https://clipquill.com/video-transcribe/ : video transcribe - https://clipquill.com/video-transcriber/ : video transcriber - https://clipquill.com/video-to-text-converter/ : video to text converter - https://clipquill.com/transcribe-audio/ : transcribe audio - https://clipquill.com/transcribe-audio-to-text/ : transcribe audio to text - https://clipquill.com/transcript-maker/ : transcript maker ## Limits and formats (as published 2026-09-18) - Input: MP4, MOV, WEBM, M4A, MP3, WAV, AAC, OGG, FLAC, OPUS, 3GP, M4B. Anything else is untested and not promised. - Output: TXT, SRT, VTT, plus one-click copy. - Limits: 30 minutes of audio and 512 MB per file. A file past either is refused up front. - Languages: the model carries 99 language tokens, but you must pick the spoken language — this build cannot detect it. English and one short Chinese clip are the only ones actually run. The Chinese run returned Traditional characters; 6 of its 97 reference characters were wrong. - No fetching from a link. Save the file to your device first. ## Machine-readable resources - Agent docs: https://clipquill.com/ai/ - Public facts (whole dataset, 19 entries): https://clipquill.com/api/agent/facts - Fact search: https://clipquill.com/api/agent/lookup?q= - OpenAPI 3.1: https://clipquill.com/openapi.json - MCP server (Streamable HTTP, anonymous): https://clipquill.com/mcp - MCP server card: https://clipquill.com/.well-known/mcp/server-card.json - API catalog (RFC 9727): https://clipquill.com/.well-known/api-catalog - Agent skill: https://clipquill.com/ai/skills/site-lookup/SKILL.md - Skills discovery index: https://clipquill.com/.well-known/agent-skills/index.json - ARD manifest: https://clipquill.com/.well-known/ai-catalog.json - Service index (DNS-AID): https://clipquill.com/ai/index.ilang - Auth status: https://clipquill.com/auth.md — under construction, no authentication exists or is required - Homepage as Markdown: request https://clipquill.com/ with `Accept: text/markdown` ## Optional - Full homepage text: https://clipquill.com/llms-full.txt (same content as https://clipquill.com/ with `Accept: text/markdown`)