Clip Scribe
Status: Live.
Your muse
Build Clip Scribe: I upload a short made-up meeting recording and get a transcript with speaker labels and keywords, plus a short made-up video clip gets masks over objects, a transparent animated sticker pack and speech bubble captions that follow each speaker, drawn in the browser. Use the voice and SAM media recipes.
Checks passed
The plan: Clip Scribe
Transcribe fake meetings and dress fake clips with masks and bubbles.
- Upload the meeting audio and the short video clip
- Name the speakers and type object words like mug and person
- Get the transcript with speaker labels and keywords
- Preview the clip with masks, stickers, and bubbles that follow speakers
See the full plan
Who it's for
Students and demo makers practicing with made up meetings and clips.
How it works
- Upload the meeting audio and the short video clip.
- Name the speakers and type object words like mug and person.
- Get the transcript with speaker labels and keywords.
- Preview the clip with masks, stickers, and bubbles that follow speakers.
- Export the transcript and save the sticker images.
Screens
- Scribe Studio. Upload audio and video, set speaker names and object words, run the jobs.
- Transcript Review. Read labeled turns, fix names, copy keywords and timings.
- Clip Canvas. Play the clip with masks, sticker pack, and speech bubbles drawn over it.
What goes in
- Audio: Short made up meeting recording. Browser converts to mono 16-bit WAV at 16 or 24 kHz. Keep under 4.5 MB or use temp upload.
- Video: Short made up MP4 or MOV clip in H.264, H.265 or MPEG-4, up to 15,000 frames. Keep under 4.5 MB or use temp upload.
- Text (optional): Optional short object words for SAM, one noun phrase per call like mug or person, plus speaker names.
What comes out
- Text: Transcript with speaker labels and turn times plus keyword list.
- Video: Preview clip with masks, speech bubbles, and stickers drawn as browser overlays.
- Image: Transparent sticker images for the pack, PNG with transparency.
Models
Model Used for Why Voice Transcribe Turn the meeting recording into text with turn timing. It is the only speech to text path for recorded files. Muse Wonder Clean the transcript, assign speaker labels, and pull out keywords. It reasons over the plain transcript and returns structured text and JSON. SAM 3.1 Find and track objects in the video clip from short noun phrases. It returns boxes and masks that the browser can draw over the clip. Muse Image Make transparent sticker images grounded in the meeting and clip. It generates the small sticker images used in the animated pack. Limits
- Speaker labels are anonymous A and B from diarization. The user names them. Plain transcript is the fallback.
- Products transcribe recorded files only, not live streams. Browser must convert to mono 16-bit WAV first.
- SAM takes one short noun phrase per call, no lists or negation, at most 16 objects.
- Video must be MP4 or MOV and masks can miss fast or hidden objects.
- Stickers are generated images sequenced in the browser, not true video generation.
- Speech bubbles are drawn in the browser from transcript timing and SAM tracks.
From your inspo
No references were given. The note asks for Clip Scribe to transcribe a made up meeting and annotate a made up video clip with masks, stickers, and captions.
Takes
- Transcribe a meeting into labeled turns with keywords
- Mask objects in a clip and track them across frames
- Speech bubble captions that follow each speaker drawn in the browser
- Transparent animated sticker pack built from generated images
Cost per run: about $0.14 (est.). Voice transcription for 5 minutes is about $0.015. Wonder cleanup and keywords is about $0.03. SAM tracking for 240 frames is about $0.048. Four sticker images are $0.04. Total is about $0.14 per run.
Recorded demo Done. Clip Scribe is done.
npm run verifyexits 0 with all 8 tests passing.It's live.