Practical guides
How to convert video to text in your browser
Video to Text turns recorded speech into editable words and timed subtitles on your device. It can help with interviews, lessons, meeting recordings and spoken tutorials. The first run downloads a Whisper model; your recording and transcript are not uploaded to a transcription server.
Choose a recording and its spoken language
Select a local video or audio file. The current tool accepts files up to 500 MiB and 30 minutes, provided your browser can decode the audio codec. MP4 with AAC audio, MP3 and WAV are useful starting points. A container extension alone does not guarantee that the audio inside can be decoded.
English is selected initially. Choose the language being spoken, or Auto detect when needed. This setting controls speech recognition; it does not request a translation. Listen to the media preview to confirm you selected the right recording.
Choose Faster, Balanced or Best
Faster is the smallest download and a practical starting point. Balanced uses a larger model. Best uses Whisper small and needs more processing time and memory. Detailed subtitle timing is optional; enabling word alignment takes additional processing time. CPU processing uses multiple threads when supported. You can also try GPU processing; speed depends on your hardware, browser and model. Try a short representative recording before using a larger tier for a long file.
Standard model weights are approximately 41 MB, 77 MB and 249 MB respectively. Best also offers GPU acceleration (~322 MB) under Processing options. This uses a separate set of weights and requires WebGPU with 16-bit floating-point support. Compatibility is checked before the accelerated weights download; unsupported devices use standard CPU processing. If the GPU fails during a run, CPU fallback may need another 249 MB download unless already cached.
Language assets, the inference engine and optional speech detection add to the first download. Cached assets can be reused while the browser retains them. These are file sizes, not memory requirements or promises of transcription accuracy. Faster remains the default, and Best GPU acceleration is optional.
Follow the transcript as it appears
Select Transcribe and keep the tab open. The tool reads audio, detects speech when enabled, loads the selected model and starts recognizing words. Draft text and timestamps may change as a section is finalized.
The subtitle panel follows new output automatically. Scroll up to read earlier text without being pulled back down; reach the bottom or select Back to latest to resume following. Pause takes effect after the current section. Stop & keep preserves the available timed segments as a partial transcript.
Review and download SRT, VTT or TXT
After completion, or after keeping a partial transcript, correct words and cue times in the editor. Check names, numbers, quiet speech and overlapping voices against the recording. The tool does not identify speakers, translate speech or burn captions into the video.
Choose SRT for a destination that requests SubRip, VTT for web caption tracks, or TXT for a readable document. Download the results before refreshing or closing the page: transcripts stay in page memory and are not saved between visits. Exporting a partial result includes only the available transcript.