Speech to timestamps
whisper.cpp emits timestamped JSON from a deployment-configurable CLI and model.
From long-form to scroll-stopping
Find the moment.
Frame the story.
Upload a short source video. ClipForge transcribes it, ranks promising moments, and renders the strongest one as a captioned vertical clip.
01 Transcribe02 Rank03 Reframe
Create a clip
Drop in your source
or
00:38▶
Bundled sample
What makes a short work?
See the complete pipeline without configuring whisper.cpp.
0%
Pipeline running
Preparing your video
FFmpeg and whisper.cpp are working locally. Keep this tab open.
Selected output
Your clip is ready
One strong idea,
ready to share.
POC ranking modeThese candidates use transparent transcript heuristics—hook words, pacing, duration, and sentence completeness. A production build would replace this seam with a semantic/ranking model.
Ranked moments
Timestamped transcript View source text
Processing stopped
Inside this proof of concept
A real media pipeline,
with explicit seams.
Rank the transcript
The POC scores self-contained windows. The heuristic is intentionally labeled—not presented as a production AI model.
Render for vertical
FFmpeg builds a 9:16 composition, burns readable captions, and produces a downloadable MP4.