| AI tool | What it does |
|---|
| cleanvoice_denoise | Remove background noise (hiss, hum, fans, traffic, room noise) from a speech recording on the user's computer — any audio or video file — offline, and save a cleaned .wav copy (16 kHz mono, optimised for speech). Use whenever the user wants to clean up, denoise or improve the audio of a voice memo, interview, podcast, lecture, call or video narration. Combine with transcription for better transcripts of noisy audio. |
| clips_make | Make a short social-media clip (MP4) from a longer video on the user's computer, offline: the part from `start` to `end` is cut out, cropped to vertical 9:16 for TikTok/Reels/Shorts (or square / original shape) and captions timed to the speech (Whisper) are burned in. Use when the user wants shorts, reels, TikToks, highlights or clips from a podcast, interview, stream, lecture or webinar. To find good moments first, get the timestamped transcript (watch_video or transcribe_transcribe), pick self-contained 20–60 s passages with a strong opening line, then call this once per clip with a clear file name for each. |
| clips_burn | Burn (hard-code) subtitles from an .srt file into a video on the user's computer, offline (FFmpeg + libass, Inter font): the captions become part of the picture, so they show on Instagram, TikTok, WhatsApp, LinkedIn and in any player. Use when the user wants captions/subtitles 'in the video', 'always visible' or translated into another language: make the .srt with transcribe_subtitles, translate or correct it yourself if asked (keep the numbering and timestamps exactly), save it, then call this. `style`: 'bottom' (classic), 'big' (large, higher up — vertical/short videos), 'box' (white on a dark box — busy backgrounds). |
| clips_audiogram | Make an audiogram (MP4) on the user's computer, offline: the part of an audio file (podcast, interview, voice note) from `start` to `end` becomes a video with the `cover` image, an animated waveform and burned-in captions timed to the speech (Whisper). Use when the user wants to share audio on Instagram, TikTok, YouTube Shorts, LinkedIn or X, which need video. To pick a moment, get the transcript first (transcribe_transcribe), choose a self-contained 20–60 s passage, then call this; one call per clip. `shape`: vertical (Reels/TikTok/Shorts), square (feeds) or landscape (YouTube/X). |
| dub_make | Dub a video into any of 31 languages (Spanish, Portuguese, French, German, Italian, Dutch, Polish, Turkish, Russian, Ukrainian, Japanese, Korean, Vietnamese, Indonesian, Hindi, Arabic, English and more) on the user's computer, offline: every caption of a translated .srt is spoken by a natural voice (Supertonic 3) at its start time, and the voice track is mixed over the original sound (`original`: kept quietly underneath, or replaced). Use when the user wants a video in another language with a voice-over ('dub this in Spanish', 'make a French version'). Steps: make subtitles with transcribe_subtitles, translate each caption yourself (keep numbering and timestamps; keep lines short enough to be spoken in their time slot — shorten rather than overrun), save the .srt, then call this. Add burned-in subtitles afterwards with clips_burn if wanted. Set `lang` to the subtitles' language. |
| dub_voiceover | Turn text into natural speech in any of 31 languages (Spanish, Portuguese, French, German, Italian, Dutch, Polish, Turkish, Russian, Ukrainian, Japanese, Korean, Vietnamese, Indonesian, Hindi, Arabic, English, …) and save an MP3 or WAV on the user's computer, offline (Supertonic 3). Use whenever the user wants a voice-over, narration or audio in a language other than English (for English, speak_speak has more voices). Write or translate the script yourself first; set `lang` to its language. Up to ~5,000 characters per call — split long texts into parts. |
| jumpcut_silence | Remove the silent pauses (dead air) from a talking video on the user's computer, offline, in one pass (FFmpeg silencedetect → exact cuts of picture and sound together). Use when the user wants jump cuts, a tighter/faster video, or 'remove the silences'. `pause`: the shortest pause that gets cut; `level`: how quiet counts as silence (noisy rooms need 'noisy'). For removing filler words, retakes or off-topic parts instead, get a timestamped transcript (watch_video or transcribe_transcribe) and call jumpcut_cut with the ranges to remove. |
| jumpcut_cut | Edit a video on the user's computer by time ranges, offline, in one exact pass (picture and sound stay in sync): `mode` remove deletes the listed parts, keep keeps only them, in time order. Use for 'edit my video by its transcript': get a timestamped transcript first (watch_video or transcribe_transcribe), decide what to cut — retakes, false starts, filler, off-topic tangents, long gaps — and pass `ranges` like '0:12-0:18, 2:40-3:05.5, 10:00-end'. Use jumpcut_silence for automatic pause removal. |
| karaoke_make | Make a karaoke video (MP4) from a song on the user's computer, offline: removes the vocals (UVR MDX-Net), recognises the sung lyrics with Whisper and burns them in as timed on-screen text over the instrumental. Use whenever the user wants a karaoke version, sing-along or lyrics video of a song. Set `language` to the song's language for better lyrics. |
| speak_speak | Turn text into natural-sounding English speech and save it as a .wav audio file on the user's computer, offline (Kokoro voices; no account, no cloud). Use this whenever the user wants a voiceover, narration, podcast intro, audiobook chapter, announcement, or wants a document/notes read aloud or 'saved as audio'. Write or clean up the script first, then pass it as `text` (up to ~20,000 characters per call; split longer documents into chapters and call once per chapter). Pick `voice` to match the request (US/UK, female/male). |
| speakers_diarize | Detect who is speaking when in an audio or video recording on the user's computer (speaker diarization, offline): returns lines like '[00:01:05 → 00:01:20] Speaker 2'. Use for meetings, interviews, podcasts and calls when the user wants speaker labels or 'who said what' — run transcribe_transcribe with timestamps as well and match the times to label each line of the transcript with its speaker. Set `speakers` if the number of people is known. |
| videomaker_make | Make a narrated video (MP4) on the user's computer, offline: the given photos/images are shown in order (each for `seconds`, looping until the narration ends), the `script` is read aloud by a natural AI voice (Kokoro, US/UK English) and captions timed to the speech are burned in (Whisper). Use whenever the user wants a video from photos with a voiceover — product promos, real-estate tours, recaps, explainers, social posts, slideshows with narration. Write the script yourself if the user only describes what they want (about 150 words per minute of video). Pick `size` for where it will be posted (landscape for YouTube, portrait for Reels/TikTok/Shorts, square for feeds). |
| vocals_split | Separate a song (any audio or video file on the user's computer) into an instrumental/karaoke track with the vocals removed and a vocals-only (acapella) track, offline with the UVR MDX-Net model. Saves two .wav files. Use whenever the user wants to remove vocals, make karaoke or a backing track, get an instrumental, or isolate/extract vocals or an acapella from a song. Takes roughly the song's length to run. |