Sashimii
Sashimii began as a test build with a single job: take a talking-head recording, get it to a clean rough cut, and get every cut in it just right.
Why We Built It
An AI agent can read a transcript and pick the good takes surprisingly well. That gets a rough cut most of the way there. The part that still eats an afternoon is the rest: a cut that clips the end of a word, a join that lands a moment too early, a pause that should have stayed. Fixing those in a full editor means hunting for each one on a timeline built for everything else.
So the first version did only this, and tried to do it fast.
What It Does
- Get your hands on the waveform. Every join shows the sound on both sides of the cut. Slide the waveform or drag the cut line, and the join plays again straight away, so you hear exactly where it lands. That hands-on control is what we found missing in the AI editing tools we tried.
- Read your video as text. The recording shows as a transcript next to the picture. Cut words are struck through, so you can see what goes and what stays at a glance.
- Check every join, one by one. Join review steps through each cut in order. Play it, accept it with ↩, nudge it a frame either way, or bring the words back.
Your own agent does the first pass from the transcript. You do the last pass by ear, and nothing slips through because you looked at every join.