How to prompt the AI video editor
Say what the finished video is, not which cuts to make. Name the outcome, the style, the length, the frame shape and what should happen to the audio — music under the dialogue, or music carrying the video alone. You never supply timestamps: Htvag Editor samples frames from every clip, reads the music's sections and beats, cuts on them, and inspects its own rendered frames before it stops.
- What to write
- The outcome, plus constraints
- Never needed
- Timestamps or cut lists
- Best giveaway
- Several clips + one track
- Always worth saying
- Length, frame shape, audio
What the editor does before it cuts
Knowing this is most of the skill. Every prompt in this guide leans on the same four things happening without being asked for.
It looks at every clip first
Before anything is placed, the editor samples frames from each video file and reads what is actually in it — the subject, the shot size, the framing, the strongest moment. That is why it can put your best shot first instead of your first file. File names and durations tell it nothing, and it does not pretend otherwise.
It reads the music, not just the file
A music track is analysed twice: where the song's sections are (intro, verse, hook, outro) so the edit starts at the main part rather than under a sparse intro, and what its tempo and downbeats are, so cut points land on the beat.
It builds a real timeline
Clips become a timeline document — in and out points, transitions, grades, captions, music on its own track with ducking under speech. You can see it, play it, and keep editing by asking for changes in plain language.
It renders frames and looks at them
After editing it renders the timeline and inspects the actual composited frames before calling the job done. An edit nobody has looked at is an edit nobody knows worked, so captions that sit off frame or a crop that cut off a head get caught and fixed.
The six things worth saying
Everything you do not name is the editor's call, which is the point — but these six change the result more than anything else you can write.
The outcome
What the finished thing is. “Make a 30-second montage” or “turn this into a talking-head clip with captions”. Not a list of cuts — it decides the cuts.
The style or the reference
A vertical Reel, a cinematic travel reel, a podcast clip, a product b-roll cut. One familiar word moves more of the edit than a paragraph of detail.
The length
A target duration, or “match the music”. Without one, it has to guess and will usually guess long. A montage of eight 3-second shots beats three 8-second ones.
The frame shape
Vertical 9:16 for Reels, Shorts and TikTok; square for a feed post; leave it landscape for YouTube. If the footage is wide and the subject is off-centre, say so — that is the crop it has to check.
What happens to the audio
Is there dialogue to protect? Should the music sit under it, or carry the video alone? This is the single most common thing people forget to say, and it decides whether the result is watchable.
The order it happened in
For event footage — an entry, a speech, greetings — say “keep the order it happened in” if the chronology matters. It usually inspects the footage and reconstructs it anyway, but naming it makes it explicit.
Any must-haves
A shot that has to open the video, a line that has to be kept, a brand title card. Everything you do not name is the editor's call — which is the point.
Looks, filters and quality
Say the look in one word and it is applied to every clip the same way — or tap one directly in the editor's Filters strip, exactly like a phone editor's filter menu. These are all the looks the editor understands, and the AI picks from the same list:
hdr
The social-media punch look: deep blacks, lifted colour and contrast, sharpened edges. What phone “HDR” filters imitate. Say “hdr look” or “make it pop like an HDR filter”.
tealorange
The blockbuster grade: shadows pushed to teal, highlights to orange. The look big action trailers and cinematic travel reels use.
goldenhour
Warm amber glow with a touch of exposure — the sunset filter. Say “golden hour look” even if you filmed at noon.
moody
Crushed blacks, muted colour, a cool tilt — the dark urban Reels look.
clean
Bright, airy, low-contrast — the clean vlog filter that lifts indoor footage out of grey.
pastel
Soft, lifted shadows and a gentle tint — the aesthetic pastel edit.
fade
The matte “faded film” look: washed blacks, low contrast, like an old photo print.
vibrant
Colour pushed forward with mild contrast — for food, travel and anything that should look juicy.
punchy
High contrast and colour for feeds and shorts — energetic without hdr's sharpened black crush.
iphone
Bright, clean, slightly warm — the default look of a recent phone camera. Useful when graded footage looks “too edited” and you want natural.
cinematic
Lifted blacks, gentle desaturation, soft contrast — the film look for travel and story edits.
warm / cool
Two ways to unify clips shot on different cameras or at different times of day.
vintage
Sepia-heavy retro — old-footage warmth and faded colour.
bw / noir
Black-and-white two ways: bw keeps gentle contrast, noir goes hard — deep blacks and strong contrast.
For quality, ask for the "sharpest possible export" or a "4K master": the editor renders at the highest resolution your footage supports and can add sharpening, which is what actually helps — Instagram and YouTube re-compress every upload, and a sharp high-resolution master survives that far better than a soft one.
Raw clips and one music track
The most common thing people bring here: several bits of footage, one song, and no idea which shot should open the video. That is exactly the case the editor is built for, so the prompt can stay short.
Here are my clips and the music. Make a 30-second vertical montage: strongest shot first and the pay-off last, cuts landing on the beat, and the music from its main part rather than the intro.
Add the clips and the track first, then paste this. It will look at every clip before it decides the order.
Order comes from content
Every clip is inspected before the sequence is decided, so the strongest shot can open the video even if it was the last file you added.
Pacing comes from the song
Tempo, downbeats and the loud and quiet sections are all measured. Clip lengths and cut points follow them instead of being rounded to whole seconds.
Length comes from you
Say a target duration or “match the music”. Eight 3-second shots hold attention far better than three 8-second ones, and the editor knows to leave good footage out.
Copy-paste prompts that work
Swap in your own footage — the shape of these is what matters. The last two are follow-ups, which are usually faster than rewriting the first prompt.
Montage from several raw clips plus one music track
Turn these clips into a travel montage cut to this music. Vertical 9:16, about 30 seconds, start at the strongest shot, keep the order the trip happened in, and use the music's main part — not the intro.
The one it is built for: it inspects every clip, picks the best few seconds of each, orders them by content, and lands the cuts on the beat.
Event reel — entry, speech, greetings
Make this event footage a 9:16 reel with the music. Keep the real order — entry first, then the speech, then the handshakes — caption the speech lines, and grade everything hdr for a punchy social look.
For rally, wedding or stage footage. A speech cut before the arrival it followed reads wrong to every viewer — the editor reconstructs the chronology from the footage.
Vertical short from one long recording
Make this a 9:16 short under 45 seconds. Open on the most interesting moment instead of the beginning, cut the dead air, and burn in captions.
Useful when the good part is 40 seconds into a 6-minute file.
Podcast or interview clip
Cut this into a 60-second podcast clip: every pause over 0.6 seconds gone, captions on, the best self-contained section only, and a title card with the topic.
It reads the transcript to find a section that stands on its own.
Talking head, tightened
Trim the silence at the start and end of every sentence and keep the whole thing under two minutes.
Say “the whole thing” when you do not want it cut down to a highlight.
A filter or a look
Grade all of this one look — hdr, punchy, cinematic, warm or iphone style — slow push-in on the still shots, and fade out at the end. Keep it landscape for YouTube.
“One look everywhere” matters — a graded clip next to an ungraded one reads as a mistake.
Extra sharpness for the export
Export this at the highest resolution you can and add sharpening so it survives Instagram's compression.
Platforms re-compress every upload; a sharpened, high-resolution master comes out the other side looking closer to what you see in the preview.
Music under speech
Lay this track under the interview, quiet enough that the voice always wins, and fade it in at the start and out at the end.
The editor ducks the music automatically wherever someone is speaking.
Captions only, nothing else changed
Add captions to what is on the timeline and change nothing else.
Follow-up prompts are fine — it edits the timeline you already have.
Fixing a result you almost like
That is close, but the first clip is too slow to start — open on the third shot instead, and make the last cut land on the music.
Describe what is wrong, not a new request. It keeps the rest of the edit.
Vague prompt, same request — sharper
None of these first versions is a bad idea. They simply leave every decision to chance.
| Instead of this | Write this | Why |
|---|---|---|
| “Make this video better” | “Make this a 30-second vertical short cut to the music, starting at the strongest moment, with captions burned in.” | “Better” is not a direction. The second one names the shape, the length, the pacing source and the treatment. |
| “Edit this to the music” | “Cut these clips to this track's beat — about 25 seconds, strongest shot first, hard cuts on the downbeats, pay-off last.” | The first is not wrong, it is just thin. Naming the sequencing rules gets you an edit instead of a slideshow. |
| “Remove the bad parts” | “Remove every pause longer than 0.5 seconds and the two takes where I restart the sentence.” | “Bad” is subjective. Thresholds and specifics are things the editor can actually act on and verify. |
| “Add music” | “Lay this track underneath, starting at the hook rather than at 0:00, ducked under the speech.” | Starting music at the file's beginning puts a sparse intro under your busiest shots — the usual reason amateur edits sound wrong. |
Getting it right without wasting credits
Add everything first, then prompt
The editor plans against what it can see. Dropping the music in after the first prompt means the first result had nothing to cut to.
One clear request per turn
“Make it vertical and cut the pauses and add captions” is fine. Three unrelated rethinks in one paragraph is where edits get muddled.
Feedback beats rewrites
“The opening is too slow, use the shot of the street” adjusts one decision. Re-sending the original prompt throws away everything it got right.
Ask for a preview before the render
Nothing is final until you export. Play the timeline, look at it, then ask for the last round of fixes — playback is free.
Prompting the video editor — questions answered
How should I prompt an AI video editor?
Describe the finished video, not the steps: what it is (a montage, a short, a podcast clip), how long it should run, what shape the frame is, and what should happen to the audio. The editor inspects your footage and the music itself, so you never need to give timestamps or tell it which clip goes where — but telling it the style and the target length changes the result more than any other detail.
Do I need to tell it which clips to use first?
No. It samples frames from every video file before placing anything, so it can order the shots by what they actually contain — opening on the strongest shot rather than on whichever file you added first. If you do have an order in mind, say so and that wins.
Does it know which clip should come first?
For event footage, usually yes — it inspects every clip and reconstructs the chronology, so the entry plays before the speech it preceded and the greetings follow the speech. If the event order genuinely matters to you, say “keep the order it happened in” in the prompt; otherwise it optimises for the strongest opening shot.
Can it apply filters like CapCut — HDR, 4K, an iPhone look?
Yes, by name: hdr, punchy, iphone, cinematic, warm, cool, vintage and bw are one-word looks the editor applies uniformly across the timeline, and it can add sharpening on top. For resolution, ask for the sharpest possible export or a 4K master — a sharp, high-resolution master survives Instagram's re-compression much better than a soft one.
Can it add memes and stickers from the internet?
Ask for the meme and it searches the web for where to download it and hands you the links — you download the file and attach it, and the editor places it on the timeline at the size, position and tilt you asked for, cut on the beat like a real meme edit. It never downloads media by itself, so nothing lands in your reel that you did not choose.
Is there a wrong way to prompt?
The main one is describing edits instead of an outcome — “cut at 0:12 and add a crossfade at 0:18” leaves it no judgement to apply. The second is leaving out the audio: if there is dialogue, say whether music should sit under it or whether the music carries the video alone.
How do I get a Reel, Short or TikTok out of landscape footage?
Ask for vertical 9:16 and say whether the subject is centred. The editor letterboxes into the vertical frame by default, can fill the bars with a blurred copy of the clip instead, and crops when the subject needs to fill the frame — but a centre crop will cut off anyone standing off-centre, so it checks the rendered frames for exactly that.
Are my videos uploaded to a server?
No. The files are read and rendered in your browser tab — the preview, the analysis and the final MP4 all happen locally. Because the media lives in the tab's memory, keep the tab open while you work: reloading or closing it means adding the files again. Very long files are the exception worth knowing about: the browser editor is comfortable up to about 15 minutes per file.
Can I keep iterating after the first edit?
Yes, and that is the fastest way to a good result. Follow-up prompts edit the timeline that already exists, so “opening shot is too slow, use the third clip” or “make the music quieter under the voice” adjusts one thing instead of rebuilding everything.
What if I do not like the whole result?
Ask for the change in plain language rather than starting over — it will re-edit in place. If a single request went the wrong way, describing the specific thing that is off (“too fast”, “the captions sit too low”, “open on the wide shot”) is more effective than a new prompt, because the rest of the edit stays intact.
Now try it on your own footage
Drop in a few clips and a track, paste one of the prompts above, and watch it decide the order for itself. Editing and previewing happen in your browser — nothing is uploaded.