AI video editor: from ten phone clips to a reel with captions

Choose and order the material: what to cut and how long a reel should be

Start by watching every clip once and sorting it into three piles: keep, maybe, bin. Phone footage from one afternoon has the same problems: the first and last second of each clip is you pressing the button, two or three clips show the same thing from almost the same angle, and at least one is a vertical clip where the phone is still turning. You need far less than you think. Ten clips of 10-20 seconds each add up to two or three minutes of footage, and the finished reel will use 30-60 seconds of it.

Why 30-60 seconds? Below 30 s there is rarely room for a beginning, a middle and a payoff; above 60 s you are asking people in a feed to commit to a video they did not search for. The platforms accept much longer uploads, but length the viewer does not need gets skipped, and the skip is what gets measured. If you have a real story (a recipe with six steps, a before and after of a renovation), 60 s is the ceiling for a reel; anything longer belongs in a landscape video for YouTube, which is a different edit.

The first 2 seconds decide whether the rest is watched. Someone scrolling sees the first frame and perhaps the first movement, and in that time decides to stay or move on. So the opening clip is not the chronological start of the day; it is the clip with the most movement, the clearest subject or the most surprising image. The establishing shot of the car park goes at the end or in the bin. A useful test: pause each candidate on its first frame. If the still image does not tell you what the reel is about, it is not your opener.

Rules for the cut list:

• One idea per clip. If a clip has two good moments 8 seconds apart, it becomes two clips in the edit.
• Three to six clips for 30 s, six to ten for 60 s. Fewer and the reel feels static; more and no shot lasts long enough to register.
• Mix distances. A wide, a medium and a close-up of the same subject read as deliberate; three mediums in a row read as an accident.
• Keep the clips with speech if you want captions. A clip with no sound cannot be captioned, and a reel where captions appear and vanish at random looks broken.

Photos can go in too. A single still held for 2-3 s between two clips is a classic way to slow the eye down before the payoff.

Automatic captions: why most people watch reels without sound

Feeds autoplay muted. People watch on the bus, in bed next to someone sleeping, in a waiting room, at their desk during a call. A reel that depends on hearing the words loses those viewers in the first second, because they see lips moving and nothing else. Captions are not an accessibility extra on a reel; they are the main channel for the words, and the audio is the bonus for the people who turn it on.

The style that works for short vertical video is word by word, often called karaoke captions: the current word is highlighted in a colour as it is spoken, and the line shows only a handful of words at a time. The reason is reading speed. A full sentence on screen is read in a second and then sits there while the audio catches up; a highlighted word pulls the eye along at the pace of the speech, so the viewer is never ahead and never behind. Two or three words per line, large type, placed in the lower-middle of the frame but above the area where the platform draws its own buttons and description.

Automatic captions come from a speech-to-text model, and speech-to-text models are reliable on ordinary words and unreliable on exactly the words you care about: names, brands, place names, product codes and numbers. 'Fourteen' becomes 'forty', a surname becomes a common noun that sounds like it, a street name becomes two unrelated words. These errors are small in a transcript and enormous on screen, because the caption is the only thing a muted viewer reads. So before you export, read the caption text once, end to end, with the audio off, and fix:

• Every proper noun: people, companies, places, products.
• Every number, unit and currency.
• Spoken filler that was transcribed faithfully ('er', 'like', 'so yeah') and reads badly as text. Deleting it from the captions is normal practice.
• Homophones the model guessed wrong: 'there' and 'their', 'weather' and 'whether'.

An honest warning: automatic captions need speech. If your clips are a montage with music and nobody talking, there is nothing to transcribe, and the right move is two or three short on-screen titles instead ('Day 1', 'The result'). Trying to caption background chatter produces nonsense. If the speech was recorded somewhere noisy (a market, a concert), expect more errors and budget time to correct them, or record a short voice-over in a quiet room and use that as the audio bed. For caption file formats and what each platform accepts, see how to add subtitles to a video for social media.

Edit video with AI: pace, colour and sound

Even when you edit video with AI, pace is your decision, not the model's: it is the length of each shot and where the cuts land relative to the audio. A dynamic reel cuts every 1.5-3 seconds and often on the beat of the music; a calm one holds shots for 4-6 seconds and cuts on a pause in the speech. Neither is better. A product demo, a sports clip or a party wants dynamic; a tutorial, a testimonial or a landscape wants calm, because the viewer needs time to read the frame. The mistake to avoid is mixing: three fast cuts followed by a 10-second hold feels like the editor fell asleep. Choose one pace for the whole reel and reserve a single deliberate change of pace for the payoff.

Transitions should be invisible. A straight cut is right nine times in ten; a short crossfade is right when time passes between two clips. Spins, zooms and page turns announce that someone discovered a transitions menu. The one exception is a quick zoom at the moment the music drops, used once.

Colour correction on phone footage should be gentle, and it should be the same across all clips. Ten clips shot across an afternoon drift: the ones in shade are blue, the ones at sunset are orange, the ones indoors are green under fluorescent light. One light grade (slightly warmer, a touch more contrast) applied to everything does more for a reel than any individual retouch, because consistency is what makes separate clips feel like one video. Avoid heavy cinematic looks on phone footage: the sensor has already compressed the highlights, and crushing the shadows further leaves a muddy image on a small screen.

Sound has two rules:

1. Even out the volume. Phone clips record at wildly different levels: one person talking close to the microphone, then a wide shot where the same person is 5 m away. Loudness normalisation brings every clip to the same perceived level so the viewer never reaches for the volume button.
2. Music under speech, not over it. If there are words, the music sits well below them, and it fades out over the last second or two instead of being chopped. An abrupt music stop at the end of a reel is the audio equivalent of a frame of black.

Shaky footage can be stabilised in software, but stabilisation works by cropping into the frame and moving the crop to counter the shake. On a vertical clip that was already tight, the crop can take off the top of a head or the subject's hands. Use it on walking shots and handheld pans; skip it where the framing is already tight; and never expect it to rescue a clip that is blurred by motion rather than merely shaking. Blur is lost information, in video as in stills, which is the same limit described in how to fix a blurry photo.

Format and branding: 9:16 for Reels and TikTok, 1:1 or 16:9 elsewhere

Format is decided by where the video will be watched, and it changes the edit, not just an export setting.

• 9:16 vertical (1080 × 1920 px) for Instagram Reels, TikTok and YouTube Shorts. Full screen on a phone, the subject can fill the frame and captions sit in the lower-middle. The platform overlays its own interface along the right edge and across the bottom, so keep anything that must be read out of roughly the bottom fifth of the frame and away from the right-hand edge.
• 1:1 square (1080 × 1080 px) for feed posts and for sites where the player is embedded in a column of text. It is a compromise: landscape clips lose their sides, vertical clips lose top and bottom.
• 16:9 landscape (1920 × 1080 px) for YouTube, websites, presentations and anything watched on a monitor or television. Here the first 2 seconds matter less and 60 s is no longer the ceiling.

Most phone clips are shot vertical, so 9:16 is the natural output. If a clip was shot landscape and must go into a vertical reel, there are two honest options: crop to the part of the frame that matters (fine if the subject is central), or show the landscape clip over a blurred, enlarged copy of itself, which keeps the whole image and still fills the frame. Stretching is never an option.

Branding on a reel is a watermark, not a billboard. A logo around a tenth of the frame width in one corner, with a margin so the platform's controls do not cover it, and your @handle in small type near the top are enough. Both should be present from the first frame to the last, because reels get screen-recorded and reposted without credit, and the watermark is the only thing that travels with the video. Pick a corner that stays clear across all your clips; top right is the usual choice because the platforms draw their own controls bottom right. The logo should be a PNG with a transparent background, in white or a single light colour, because it will sit over footage of unpredictable brightness.

Caption colour belongs to the brand too. The highlight on the active word is the one element that repeats in every reel you publish; using your brand colour for it (as long as it contrasts with both dark and light footage) makes a series of reels recognisable in a feed before a single logo is read.

Step by step in Airpic: make reels with AI, from upload to a 1080p export

Airpic's AI video editor is built around the sequence above: it asks the questions first, proposes an edit, lets you correct it in plain language, and only charges when you export. If you want to make reels with AI rather than learn an editor, these are the steps as they appear on screen.

1. Files. Drop up to 10 videos or photos (MP4, MOV, WebM, JPG, PNG, WebP; up to 1 GB each). Each file shows as a tile with its length, orientation and whether it has sound. Use 'Move up' and 'Move down' to put your opener first, and remove anything from the bin pile. 'Add music (optional)' takes an audio file; 'Add logo (optional)' takes your PNG. Then press 'Analyse and continue'. The speech in every clip is transcribed on GPUs in the EU, free, and the first time the engine can take a minute or two to start. Clips with no speech are flagged ('no speech detected') so you know which ones will have no captions before you commit to anything.
2. Brief. A panel appears with every setting from this article, each with a recommended value already selected, and nothing is edited until you confirm. Format: 9:16 vertical for a reel. Target length: 30 or 60 s. Pace: Dynamic or Calm. A free-text box for what the video is about or how it should end ('highlights of the launch, end with the shop address'). Automatic captions: on, style 'Word by word karaoke', active word colour set to your brand colour, size Normal or Large. Below it, the full caption text is shown and editable: this is where you fix names and numbers, and what you type is exactly what appears on screen. Colour: Natural or Warm for phone footage. Improvements on export: Stabilise, Reduce noise, Even out volume (tick the last one for anything with speech). Brand: @handle and the logo corner. Music: volume and 'Fades out at the end'. Then press 'Confirm and propose an edit'.
3. Edit. The AI receives the transcript, your brief and each clip's details (length, orientation, whether it has sound), not the videos themselves, and returns a proposal: order, cut points, transitions, titles and captions. It plays at once in a free preview in your browser, with a 'Cuts' strip showing each clip and the total length. Under it is 'Adjust by chatting': type what you would change ('remove the second clip', 'faster at the start', 'bigger captions', 'add a title at the start saying Open Saturday') or use the one-click chips, and the preview updates. Change it as many times as you like; the preview never costs anything. 'Edit captions' reopens the text if you spot an error while watching, and 'Back to the brief' changes format or pace without re-uploading.
4. Export. 'Export MP4 1080p' renders the final video on servers in the EU with the real colour grade, stabilisation, loudness normalisation and burnt-in captions. Credits are taken before the render and refunded automatically if it fails. The export takes about as long as the video, or a little more: in our test on 16 October 2026, a 15-second edit at 1920 × 1080 px exported in 40 seconds. Download the MP4 and upload it to the platform directly, not through a messaging app, which recompresses it.

Two related tools cover the cases this editor does not. If your source is one long video (a talk, a podcast, a live stream) rather than a folder of clips, make shorts from a video finds the strongest moments, reframes them to 9:16 following the speaker's face and adds the same word-by-word captions. If you only need the words (for a post, a description or your own subtitle file), transcribe video to text returns TXT, SRT, VTT and JSON in one ZIP. And if the finished reel is going on a screen bigger than a phone, upscale video to 4K takes a clip of up to 20 seconds to 1080p or 4K.

What the editor is not: a multitrack timeline with keyframes, masks and audio ducking curves. If your reel needs picture-in-picture, a green screen or twelve cuts synced frame-accurately to a drum pattern, learn a professional editor; this one is for the ten-clips-to-one-reel job, which is most of them. Your files are deleted when the job finishes and the export one hour after you download it (24 hours at most), and nothing is used for training.

FAQ

Keep reading