Upload a video or audio file and get back SRT and VTT subtitles plus an MP4 with captions burned in. Ask for translated subtitles in the same message, no API key needed. Cues are re-split by line length, characters per second and on-screen time; if a translation's cue count or timecodes drift from the original, that file is not shipped. Runs in the Capafy cloud, no install. Measured: a 2:25 clip transcribed in 9 s, burned in 29 s on 2 CPUs. The weekly plan starts with a free trial period.
Removes burned-in captions and subtitles from short videos (MP4, MOV, WebM) and returns a cleaned MP4 with the original audio.
Gets word-level STT with faster-whisper and re-splits cues by broadcast-style rules (max 2 lines, chars per line, CPS, min/max duration). In a 41.78 s sample, 7 raw whisper segments (up to 51 chars on one line) became 12 cues (up to 18 chars, 2 lines); STT word accuracy was 98.4%. The agent fills translation worksheets, so no API key; a script checks cue count and timecodes against the source and writes nothing if any differ. Burns in outline/box subtitles; QC: 0 issues in 3 languages.
Turn an uploaded MP4, its timed SRT subtitles, and up to three chosen time ranges into vertical clips. Each MP4 keeps the original audio and includes burned-in captions. Get a matching SRT file and source timestamps for each clip.
Downloadable toolkit, not a cloud service. Runs on your own computer inside Claude Code and turns your content into narrated, captioned vertical Shorts (MP4). Auto-upload to YouTube needs your own Google Cloud OAuth setup.
Check your .srt or .vtt captions before you post. An offline script finds exact cues with reading speed, line length, overlaps, timing errors, broken characters and script mismatches, fixes the mechanical problems without changing a word, and re-checks. It cannot hear audio or test platform acceptance. Download package: bring your own AI agent and Python 3.
Paste the caption you intended plus what the platform actually saved, and get a character-level diff that catches silently corrupted hashtags, truncation, and stripped formatting — plus a paste-safe caption block built to survive the editor that mangled it.
Give it a blog post or article and it writes a script with a hook and 5 to 8 beats. Two local Python scripts do the rest: edge-tts voices each sentence, then Pillow and ffmpeg combine a Ken Burns background (your image or a gradient), keyword-highlight captions and a progress bar into a 1080x1920 video. In our sample, an 870-character Korean article became a 48.3-second video. No API key needed; with BGM, sidechain ducking is automatic.