How to write a good prompt for Cut Express.
Your AI makes every decision in a Cut Express video: which moments to keep, where to cut, what the captions say, which sound lands where. It decides well when it knows what you want. This page is what to tell it, in the order it matters.
The short version
- The story and the length: what happens, in one sentence, and how many seconds.
- The format: 9:16 for Reels, Shorts and TikTok, 16:9 for YouTube.
- What you uploaded: what each file is, and what to take from it.
- The hook: what the first two seconds show and say.
- Captions, look, sound and pace, in a line each.
- "Show me the timeline before you render."
- Then notes: one change per message, each one a new version.
That is seven lines. Everything below is why each one matters, with a vague prompt next to a good one.
Say the story and the length
The story and the length are the two things Cut Express needs before anything else; for a new video, your AI starts with brief, which asks for them when they're missing. "A reel from my trip" has no story. "Lisbon in 30 seconds: arrival, the trams, the food, sunset at the miradouro" does, and your AI can cut to it.
make a video from my clips
Make a 30-second reel of my weekend in Lisbon from my clips: arrival, the trams, the food, and the sunset at the miradouro last.
Name the format and where it goes
Say 9:16 or 16:9, and the platform. The platform tells your AI how fast to cut and where the app's buttons cover the picture, so captions and texts stay out of the way. Need both? Ask for 9:16 first, then "the same timeline again as 16:9": it's the next version of the same video, so every note carries over.
Say what you uploaded
Your AI sees your files as contact sheets (frames labelled with their time) and, when it needs the words, a transcript. It still helps to say what each file is: "clip 1 is the intro, the long one is the full talk, the MP3 is my music". For a long recording, say which part matters, or ask for moments (below).
Put the hook first
The first two seconds decide whether anyone watches the rest. Tell your AI what opens the video: the most surprising line, the result before the process, the best shot. If you have a hook line, give it in quotes; it can be said, shown as a text, or both.
start with an intro
Open on the line "I almost quit at 2 AM" even though I say it at minute 4, with that line as a big text at the top for the first 2 seconds.
Choose the captions
Captions come from the speech, word by word, timed to the real audio. Say the style (pop, karaoke, boxed, bold, clean or minimal, or "you pick"), the highlight color, how many words per line and where they sit. Two or three words per line reads best on a phone. If the video has text on screen, say where the captions should go so they never cover it.
Describe the look
A look is a color grade. Describe it by mood ("warm and soft", "crisp and bright", "teal and orange", "black and white") and your AI picks from the looks and adjusts contrast, saturation or grain. One look over every clip makes footage from different cameras feel like one film.
Say what the sound should do
Sound is half the edit. Say whether to keep the clips' own sound, which music (a track you uploaded, or a mood for the stock music search) and whether it ducks under the voice. Ask for a voice-over from a script (say the voice you want) and sound effects where they belong: a whoosh on transitions, an impact on the reveal. Voices and made sound effects spend credits; the built-in effects are free.
Set the pace
"Fast cuts on the beat", "calm, let the shots breathe", "speed ramp on the jump, slow on the landing", "punch-in zooms on the key words". Pace is the difference between a video that feels edited and one that feels assembled.
make it engaging
Hard cuts on the beat for the first half, then slow it down for the last shot. Punch in on the words "twice as fast". No transitions except one fade at the end.
Ask to see the timeline before it renders
Add "show me the timeline before you render" to any prompt. Your AI lists the scenes with their times, the texts and the sound before anything spends render minutes. For long footage, ask for the list of moments first and pick from it. Fixing a line on a list is free; fixing a finished render costs another render.
Iterate with versions
Every render is a version of the same video, and its timeline is kept. Give notes like you would to an editor, one or two at a time: "make scene 3 shorter", "captions in yellow", "music lower under the voice", "swap the hook for the laugh at 1:12". Each note renders the next version; the earlier ones stay in your account, so a note that doesn't work costs nothing.
redo it, it's not good
v2: cut scene 3 to 2 seconds, move the caption up so it doesn't cover my face, and bring the music down by half under my voice. Keep everything else.
Long footage: ask for moments
For a podcast, a stream, gameplay or a vlog, ask your AI to inspect it with moments: Cut Express finds the stretches where something happens and ranks them, with a frame sheet for each. Say what a good moment is for you ("the clutch rounds", "where we argue about pricing") and how many clips you want. Your AI then zooms into each stretch before choosing the in and out points, so it never guesses a timestamp.
find the best parts of my podcast
Inspect my 58-minute episode with moments and a transcript. Find 5 stand-alone moments of 30 to 45 seconds where we disagree about something. List them with their times and the first line, then wait for my pick.
Recreate a video you like
To copy a reel's style, ask for the recreate panel: one box for the video to copy, one for your clips. Your AI breaks the reference down (its shots, their lengths, the movement, the beats, and its first three seconds frame by frame), then matches each shot with one of yours and keeps its pace, its texts and its caption look. The reference itself is never cut into your video. See Recreate a reel you like.
Motion graphics and motion pieces
Motion graphics (animated titles, charts, counters, kinetic text cued on your words, your app on a screen in the footage) go on top of a normal edit: say what each graphic shows and when. Give real numbers; your AI animates what you give it and never makes data up.
A motion piece (a logo reveal, a launch film, a title sequence) is directed: it starts with a brief, then a script you approve, a storyboard, a design board image you approve, and previews of every few seconds before the render. Answer the brief's questions (the story, the length, who it's for, how it should feel, your brand colors, references you like) and approve each step before the next. See Directed motion piece.
Before and after: a whole prompt
make a promo video for my app
Make a 20-second 9:16 promo for Lumen, a budgeting app for freelancers, from the screen recording I uploaded. Hook: open on the dashboard with the text "Taxes, done in 3 taps". Then the three screens: new invoice, auto tax estimate, the payment arriving. Punch in on each tap. Voice: a calm female voice, this script: "..." Captions from the voice, bold, highlight #FFD23F. Soft lo-fi music ducked under the voice, a click on each tap. End card: "Try it free at lumen.app". Show me the timeline before you render.
Common mistakes
- No length. Without it, your AI guesses. Give seconds.
- Ten notes at once. Some get lost. Give two or three per version.
- Asking for facts it doesn't have. Prices, numbers, names and quotes come from you; your AI won't invent them.
- Rendering before looking. Ask for the timeline or the list of moments first.
- A long voice before the script is final. Voices spend credits; settle the words first.
Ready? Browse the prompt library and start from the one closest to your video.