Editing real footage
Cut your footage from the words inside it
Whisper transcribes each take to the word. Kamply packs that into phrases with second ranges, you say which phrases stay, and the cut list lands as a JSON file that ffmpeg renders on your own machine.
6 ranges x 5.0s = 30.0s
- 5 crossfades x 0.5s = 2.5s
= 27.5s
Figures read from videos/sample-30s-edl/edit.edl.json, the worked example that ships with Kamply.
What it actually does
Kamply is a Mac desktop app. You give it your takes, it runs Whisper over each one with word timestamps, and packs the result into a plain phrase list where every line carries its start and end in seconds. You say which phrases stay. That becomes edit.edl.json, a cut list you can open and read, and ffmpeg renders it: each range extracted with the colour grade baked in, joined by hard cut or crossfade, subtitles burned last in sentence case. The MP4 lands in your project folder.
What this is not
The timeline Kamply shows you is read only. It plays the cut back with separate lanes for film, text and audio, and nothing in it can be dragged or trimmed. Kamply does not generate footage, does not do avatars or lip-sync, and does not upload the finished file anywhere.
How it runs, start to finish
Three steps, in the order the app actually does them.
Whisper reads every take
Each take runs through Whisper with word timestamps on. The packer turns that word JSON into a plain phrase list, breaking on any silence of half a second or more, with every phrase prefixed by its start and end in seconds. That file is what gets read to choose cuts, so you are picking moments by reading them.
The cut list is a file you can open
Your choices land in edit.edl.json: a map of source clips, then one range per cut with a start and an end. A range can carry a crossfade measured in frames, or a slow Ken Burns zoom for silent stock. One colour grade covers the whole cut and one subtitle file rides along. Change a number and re-render.
ffmpeg renders it, then the gates read the file
Each range is extracted with the grade and a 30 ms audio micro-fade baked into both edges, joined losslessly or through a single crossfade graph, and the subtitles are burned last. The automated check then runs the same 18 numbered gates a Remotion-built reel faces, six of which read the finished MP4 itself: audio rate, freeze, cadence, gaps, loudness and the timeline gate.
The numbers, and where they come from
Every figure on this page traces to a published source or to Kamply's own measurements. Nothing is modelled.
- Phrase break
- 0.5 s of silence
- The packer starts a new phrase at any gap of half a second or more, then prefixes it with its second range.
- Render output
- 24 fps h264, aac at 48 kHz
- The renderer's fixed output. 48 kHz is pinned through the audio master too, after a loudnorm pass once emitted 96 kHz and social players drifted out of sync.
- Micro-fade at every cut
- 30 ms on both edges
- Baked into each segment so no cut pops. Loudness is handled separately, in the audio master.
- Final loudness
- -14 LUFS, -1.5 dBTP
- One loudnorm pass in the audio master, the social delivery standard, checked again by the gate.
- Caption safe zone
- MarginV 90
- Lifts captions clear of the bottom quarter that TikTok, Reels and Shorts cover with their own UI.
- Price
- $199 a month
- One plan: $69 weekly, $199 monthly, $1,990 yearly. Founding rates end 2026-09-15.
Questions people actually ask
If yours is not here, ask it in the walkthrough. The form below reaches a human.
Is there a timeline I can scrub and drag?
You can scrub it. The Studio view plays the render above a zoomable timeline with a ruler, cut markers and a playhead, and every lane on it is drawn from a file on disk. Dragging is the part that is missing. Claude in the embedded terminal is the only thing that writes files, so you change a cut by changing the cut list. If dragging clips is the job, this is the wrong tool.
Can I see the footage before committing to a cut?
Yes, as a picture rather than a player. Point the timeline tool at a take and a time range and it renders one PNG: ten filmstrip frames, an RMS waveform underneath, and the spoken words labelled along it. Cut points get chosen against the audio shape. When you want to watch the cut rather than read it, a lower-quality preview render runs before the full one.
Can it add lower thirds, animated titles or effects?
Not on this path. The ffmpeg cut does three things: cut, grade, subtitles. Nothing draws on top of the frame. A video that needs an animated lower third becomes a Hybrid instead, where Remotion owns the timeline and your footage rides inside the composition as an asset, so the overlay is built there.
Does it know who is speaking in an interview?
No. It runs Whisper, which gives word timestamps but not speaker labels, so a two-person interview loses who said what. You would mark the speaker turns yourself. Single-speaker footage, a founder to camera or a product demo, is what the path is built for.
Will it touch my original files?
Never. Raw takes keep their original filenames and stay put. Every derived file is a new file whose name records the treatment, so a graded segment and a sentence-cased subtitle copy sit beside the source rather than replacing it. The transcripts are cached separately and the cut list references takes by id and path.
What does it not do to the picture or the sound?
No upscaling, no stabilisation, no noise reduction, no de-reverb, no speed ramps and no reverse. The only join it writes is a hard cut or a crossfade measured in frames; there is no effects library behind it. HDR footage from an iPhone or a mirrorless camera is tone mapped to clean Rec.709 first, which is the one correction it makes to the picture on its own.
Bring one brief. Watch your first campaign take shape.
We build the first direction with your brand during the walkthrough. You leave knowing exactly what Kamply would make next.