AI B-Roll Editor: From Messy Footage to a Finished Video
Do it with the AgentNo footage or script yet?
B-roll Cut is an AI B-roll editor: it watches every clip in an unsorted folder, puts the right shot under every line of your script, and cuts one finished MP4 with voiceover and captions. Paste the script or upload your own recording, then drop in the whole folder.
Last updated
Four steps to a finished video
- 1
Upload clips
Any number of videos (.mp4, .mov, .m4v) or photos (.jpg, .png, .webp, .heic). Each photo becomes a 4 s clip with a slow zoom.
- 2
Add the script
Paste the voiceover script, in English or Chinese, up to 30 minutes of speech (8 minutes on the free cut).
- 3
AI picks the shots
AI reads every clip and picks a shot for each phrase of the script.
- 4
Render and download
Render one MP4 with the voiceover and burned-in captions, and download it.
Who uses it
YouTube explainers and essays
Cover a long narration with stock or archive B-roll without scrubbing through hundreds of clips by hand.
Product and brand videos
Match your own product footage to a launch script, voiced and captioned in one render.
Short vertical videos
Cut a 9:16 video for TikTok, Reels or Shorts with fast shots and captions that viewers can read on mute.
Chinese talking-point videos
Paste a Chinese voiceover script and get footage cut to every line, with Chinese captions burned in.
What a finished video costs
- Every registered account gets one free cut: a script of up to 8 minutes (8,000 characters), any number of clips, and up to 3 AI voices or your own recording. After that the editor uses purchased credits.
- Credits are charged step by step: reading clips (per batch of 8 clips), matching the script (per part of the script) and rendering (per minute of video).
- The editor shows the credits still needed to finish before you start, and asks you to confirm.
- A step that fails is refunded, and the work already done is kept, so a retry costs only what is left.
More AI video tools
AI Video Ad Generator
Turn a product brief into a finished video ad with voiceover and music.
AI Video Generator
Generate a single video clip from a prompt or an image.
AI Video Templates
Turn one sentence into a vertical promo, logo intro or chart video.
TikTok Carousel Maker
Turn one topic into a TikTok photo carousel with AI copy and photos.
The Complete Guide to AI B-Roll Editor
What B-Roll Is and Why Cutting It by Hand Takes So Long
B-roll is the supporting footage that plays over a voiceover: the street scene behind a sentence about city life, the close-up of hands on a keyboard behind a line about remote work. A-roll carries the story in sound, and B-roll carries it in pictures. A good cutaway changes when the words change, so the viewer always sees what they hear.
Cutting B-roll by hand is slow because the hard part is search, not editing. For every phrase you scrub through clip after clip, find the second that fits, trim it and move on. A three-minute narration with fast cuts can need a few hundred shots, and a library of several hundred clips turns each search into minutes. The editor does the searching and the trimming for you and leaves the judgment calls visible in a timeline you can check.
Preparing a B-Roll Library That Matches Well
The match can only be as good as the footage you give it. Upload clips that cover the topics your script talks about, with several options for the ideas that come up often. Short, clearly shot clips with one subject each are easier to describe than long handheld takes that wander between scenes.
Each clip can be an .mp4, .mov or .m4v file of any size, or a photo that becomes a 4 s clip with a slow zoom, and there is no limit on the number of clips in a project. Before upload, your browser compresses heavy footage (for example 4K HEVC from an iPhone) to a silent 1080p H.264 copy, which uploads many times faster and looks the same in the cut. Your browser also draws a contact sheet of it, about one frame every half second and at most 24 frames, so the AI reads your clips from those frames instead of waiting for a server to decode every file. If an iPhone .mov file fails, it is usually HEVC: try Safari, or export the clip as H.264 MP4. For very large libraries we recommend the free desktop app for Mac and Windows, because the web editor gets slow with very many or very large files.
Uploaded clips and their contact sheets are kept until you delete the project, so you can come back, change the script and cut again at any time.
Writing a Voiceover Script for B-Roll Matching
Paste the script you will record, in English or Chinese, up to about 30 minutes of speech (30,000 characters at most; the free cut takes up to 8 minutes and 8,000 characters). The editor splits it into caption-sized phrases and estimates a time for every word: English by the length of each word, Chinese by the character, with short pauses at commas and longer ones at the end of a sentence. Each phrase becomes one caption and one beat of the picture.
Concrete words give the best B-roll. A line such as "Coffee beans roll out of the roaster" points at a picture, while "Quality matters to us" could match almost anything. When a phrase is abstract, add a visual detail to the script or make sure the library holds footage for it. Phrases with no match get a filler shot, which the timeline marks so you can fix it.
How AI Chooses and Cuts Every B-Roll Shot
Matching runs in two judgments and one fixed step. First, AI reads every clip once, in batches of 8, and writes a short description and tags for each part of it. Second, AI reads that footage index together with your script, about 30 seconds of phrases at a time, and ranks the parts of your footage that fit each phrase best.
The plan comes from AI, the cutting from deterministic code. AI groups the phrases into story beats and gives each beat a short shot list that moves from a wide establishing shot to medium shots and close-ups of the same scene, so the pictures tell the story even on mute. It prefers steady footage over shaky handheld takes and may cut back to the same clip at another size. The assembler then sets every shot to the pace you chose (fast, medium or slow, shots of about 1.5 to 4 seconds), cuts it on the 30 fps frame grid, avoids jump cuts, and only falls back to the least used clip as a filler when nothing fits.
The pace shapes the plan, so a new pace matches the footage again and the match is charged again. Adding or removing clips, editing the script or notes, or changing the voice also clears the match, and the editor tells you before it does.
Rendering the Video and Adding Your Voiceover
Check the timeline phrase by phrase: every row lists its shots with the clip name, the source time and the length. When it reads well, render one MP4 at 1920 x 1080 for 16:9 or 1080 x 1920 for 9:16, with the captions burned in. Long videos render in parts of about a minute, and a render that runs again keeps every part whose input did not change.
An AI voice reads your script, and every caption and shot is timed to the moment the voice says it, so the finished cut carries the voiceover under the picture. English voices are available for now. The finished video is also saved in My Works.
Where It Fits in Your Workflow
YouTube explainers and video essays are the classic B-roll job: a narration of several minutes needs fresh pictures every second or two, and an archive of footage grows faster than anyone can remember it. Paste the narration, upload the folder for that episode, and the editor gives you a first cut to refine instead of a blank timeline. Product teams do the same with launch footage, matching a demo script to their own footage of the product in use.
Short vertical videos need a different rhythm. Pick 9:16 and the fast pace, and the shot changes every second or two, which keeps viewers watching on mute while the burned-in captions carry the message. For a calmer documentary feel, choose the slow pace so each shot has room to breathe. Chinese talking-point videos work the same way, timed by the character, with captions that break where a Chinese reader expects them.
Whatever the format, treat the first cut as a draft. Read the filler count, upload footage for the phrases that need it and match again. A few extra clips often turn a timeline full of fillers into one where every shot matches its words.
Make an AI Video From Your Own Clips
People use the words "AI video" for two different things. One kind of tool generates new footage from a text prompt. The other kind takes real footage and edits it for you. This editor is the second kind: when you make an AI video from clips here, every shot in the picture comes from a file you uploaded. If you want new footage from a prompt instead, use the AI video generator.
The AI does the editor's work, not the camera's work. It reads every clip and notes, for each part, what it shows, a few tags, the shot size from wide to detail, the camera motion and the scene. It uses shaky footage only when nothing steady fits. Then it reads your script, matches a shot to each line, trims it to the pace and adds the voiceover and the captions, so you get an AI video from your clips that you can check shot by shot in the timeline.
To start, create a project and pick 16:9 or 9:16 before you add footage, because your browser prepares each clip for that frame. Drop in the clips, then paste a script or upload your own recording: a recording is transcribed, so it needs no script. Start the match, read the timeline and render. The cut continues on the server after you close the page, and an email tells you when it is done or when it failed.
Some ideas: a faceless YouTube channel, where you write the narration and upload the footage of each episode; your own long videos as Shorts, with 9:16 and the fast pace; an explainer built from a folder of archive footage; a product video matched to a demo script. The same project covers the YouTube workflow from outline to upload, and it joins many clips in the order of a story.
Make AI Videos Using YouTube Clips the Right Way
Many people look for a way to make AI videos using YouTube clips. The editor does not download anything from YouTube or any other site. It only edits the clips you upload. So the real question is where your clips come from, and whether you have the right to use them.
Good sources are the original files of the videos on your own channel, your own shoots, and stock or archive clips you hold a license for. Reusing another creator's video needs that creator's permission, so use only footage you own or have the right to publish.
When the cut is done, you download one MP4 and publish it yourself: the editor does not post to YouTube. To keep working in your own editor, "Export timeline" downloads a free ZIP with the cut as an editable timeline for Premiere, DaVinci or Final Cut, with every shot on its own, the voiceover, the music and the captions.
What Matching Costs and How to Spend Less
After the free cut, the editor needs purchased credits, and each step is charged on its own. Reading clips costs 1 credit per batch of 8 clips. Matching is priced per part of the script and depends on how large the footage index is; the first part costs the most, and the parts after it reuse that work and cost less. Rendering costs 2 credits for each part of about a minute, plus 1 credit to join the parts. The editor shows what is still needed before you confirm.
To spend less, upload only clips you might really use, and settle the script and the pace before you match, because a change to either one matches again. A failed step is refunded and the finished work stays, so a retry pays only for what is left. When you need new footage rather than a cut, make a clip with the AI video generator, and for a finished product ad with voiceover and music use the AI video ad generator.