Every AI edit looks the same.
This one looks like you cut it.
It transcribes your footage, cuts the dead air, finds the takes worth keeping, and captions them. Then it does all of it the way you would have — and every cut comes with a receipt.
We read one of your videos by hand and send back your taste.md · no spam
Same footage, same instruction — only the taste.md differs. The first two are two schools; the vertical is the same school changing platform — its segments don't even line up, because they were re-chosen, not shortened.
Illustrative demo — every duration here gets replaced by a real timeline export.
an AI editor whose cuts don't all look alike
You've had this night.
The postable night.
47 minutes of footage. 90 seconds worth posting. You know exactly which ninety — finding them is the job you didn't sign up for.
The three-verticals night.
One good take. Three platforms. Three different first-three-seconds.
The missing-shot night.
The cut needs an opening you never filmed, a chart you never made, music you can't license.
The not-mine night.
So you tried an AI editor. Fast. Clean. And… someone else's video. Transitions you'd never touch, captions in a stranger's voice. Undoing someone else's taste took longer than cutting your own.
It's not slop anymore. It's clean, it's fast — and it all looks the same. Every AI editor ships with one set of judgments: what counts as a highlight, how long a pause breathes, where B-roll lands. You can pick their presets. You can't edit their judgment. A hundred videos cut by the same tool look like they were cut by the same person — because they were.
their receipt
Removed 12 fillers · 3 repeated takesFound 4 highlights
Done ✓ Proof that it worked.
a tastecut receipt
cut interval 1.7–2.8s → rewrote 7 cuts · "vlog03" 00:12zoom transitions: never → skipped 3 · "vlog01" 02:41
2 rules below 0.6 → asked first Proof that it's yours — every move traced to the clip it learned from.
Their receipt proves it worked. Ours proves it's yours.
Same request. Two editors.
without taste
after it learns you
The difference isn't the model. It's who it listens to first.
AI that learns your editing style — taste.md
It studies you first.
Give it three videos you've already made. It goes through them the way an apprentice would — shot by shot, cut by cut.
Live React components from the product itself — not screenshots. The buttons are the real ones; press one.
- Read it. Prose, not weights — each line says what you do, how strictly (always / usually / will ask), and where it learned it ("vlog03", 00:12).
- Edit it. It's your file. Rewrite any line in plain words.
- Switch any line off. Takes effect next cut; never rewrites the last one.
No rules to write. It writes them — you veto.
It starts before you ask
Drop footage in and it already has a starting point — read off the shape of your clips. Say where it’s going and it sharpens. Nothing to pick, nothing to fill in.
Change what you disagree with
The change itself is a preference — it's recorded on the spot.
✦ learned: B-roll lands at sentence ends · alwaysFeed it your old videos
Everything above is a starting point. This is the one path to preferences that are actually yours: it measures your cuts, line by line, for you to confirm — and a habit two of your videos agree on outranks anything it guessed.
A template is a choice you can't argue with. This one starts wherever your footage points and ends as your taste — because it remembers every argument you win.
AI video editing — transcribe, cut, caption, reframe, generate
It cuts what you shot. It makes what you didn't.
Before it can cut like you, it has to cut. Everything below is one sentence away — the same chat box, no timeline required.
- Transcribes everything, then cuts by text. Delete a sentence in the transcript; the timeline follows.transcribe · text-edit
- Kills the dead air. Filler words, restarts, the three seconds where you lost your place.7 pauses cut · 0.38s
- Finds the moments worth keeping. It reads the whole take and proposes the cut, instead of asking you to scrub.highlights → first cut
- Captions, in your style. Generated from the transcript, so they're right on the first pass.captions · styled
- Reframes for vertical. One horizontal take becomes a 9:16 cut without recomposing every shot by hand.16:9 → 9:16
- Renders and exports. The finished file, not a project you still have to open somewhere else.render → mp4
And when the cut needs something you never shot — it makes that too.
The opening you never filmed
A generated establishing shot — paced to your cut interval, not a stock rhythm.
gen shot → timeline · receipt attachedThe chart you never made
Motion graphics set in your caption style — described in a sentence, no keyframes.
MG card → timeline · receipt attachedThe score you couldn't license
Music at the exact length, tucked under your voice the way you always mix it.
score → timeline · ducked −12dBGenerated material obeys the same file as everything else. It's footage, not effects — it lands on the timeline with a receipt, next to the clips you shot.
explainable AI video editing, with receipts
It knows what it doesn't know.
Confidence is capped
One video can never teach it past 0.6. Style needs witnesses.
It audits its own cut
After every edit it checks itself against your file and flags what drifted.
◆ 2 cuts drift from your taste — jump to the first →Unchecked is a dash
If it didn't verify something, it says so. A dash, never a checkmark. And every change reverts alone.
Manual never leaves. Every AI change is an edit you could have made — on a timeline you can always touch.
A tool that says "I'm not sure" is the only kind you can believe when it says "this is you."
taste.md is sentences, not weights.
Every line says what you do, how strictly it applies, and where it learned it. You can argue with a sentence.
MCP built in.
Cut from Claude Code or your agent of choice.
The demo on this page started as one 3:20 take. It became a 1:38 fast cut, a 2:44 slow cut, and a 0:45 vertical — same footage, same instruction, different taste.md. Every duration on this page is a real timeline export.
Send us a video. Get your taste.md back.
Leave your email and we’ll reply with where to send one video you’ve already made. We read it by hand and send back your taste.md — the file, not a sales pitch. Yours to keep, plain text, whether or not you ever use TasteCut.
What does "learns my style" actually mean?
Feed it videos you already made. It measures your cut points, the interval you keep between cuts, what you never do — and writes each habit as a sentence in taste.md, with a confidence score and the exact clip it came from. Low-confidence habits are marked will ask first.
Why do AI-edited videos all look the same?
Every AI editor ships with one set of judgments — what counts as a highlight, how long a pause breathes, where B-roll lands. You can pick their presets, but you can't edit their judgment, so every output carries the vendor's taste. TasteCut replaces that judgment with a file you own.
I don't have old videos to feed it. Does it still work?
Yes. Drop footage in and it reads a starting point off the clips themselves — aspect, length, how fast you talk; say where it’s going and it sharpens. Feeding old videos later is what turns that starting point into preferences that are actually yours.
Can I still edit manually?
Always. The timeline never leaves. Every AI change is an edit you could have made yourself, each one revertable on its own.
Is my footage used for training?
No. We don’t train anything on your footage or your taste.md. Editing does route material through AI providers — transcription, understanding, generation — because that’s how the work gets done; we send only what the edit needs, and keep nothing from it for training.