CapCut: subtitles and AI B-roll (2026)
CapCut targets short social video. This flow assumes you already have a 30–90 second talking-head clip (or voice with a static frame) in vertical 9:16. In about 22 minutes you leave with synced subtitles, a style that reads on mobile and at least one B-roll cut — from **Stock materials** or a short clip you import from Runway/Kling if you generated it elsewhere. We are not cutting a documentary: it is the «caption + one supporting shot» pattern you repeat on every short.
The problem it solves
Many people export without reviewing Auto captions and ship misspelled names or out-of-sync lines. Another failure mode: random B-roll that does not illustrate the line. Or loud music during caption generation and then blaming the ASR. CapCut automates the first pass; you validate text, pacing and whether the visual cut matches the script.
What you will learn
- Create a vertical project and import A-roll
- Generate Auto captions with the correct spoken language
- Fix subtitle blocks and basic timing
- Apply a caption style readable on a small screen
- Find B-roll in Stock materials or import a short AI clip
- Place B-roll in a script gap without burying the message
- Export MP4 and check sync with headphones
Before you start
- An account at capcut.com (web or desktop app; the free tier is often enough for one short)
- A 9:16 A-roll video with clear voice, 30–90 s (MP4 or MOV)
- 22 minutes and headphones to review captions and cuts
Steps 1
New project and import vertical A-roll
Open **capcut.com** (web editor) or the desktop app and sign in. Create a **new project** and set **9:16** aspect ratio if the wizard asks (you can also change it in project settings). Use **Import** and upload your talking head. Drag the clip to the main video track. Trim dead air at start and end with **split/trim** on the timeline — keep only what you would publish. Confirm the **audio** track is not muted and voice level is audible; without clear audio, Auto captions have nothing to transcribe (CapCut official docs).
What you should see: One clip on a 9:16 timeline, playhead at the start of speech, voice waveform visible.
Steps 2
Generate Auto captions
In the side panel, open **Captions** or **Text** (exact label varies on web vs desktop). Choose **Auto captions** / **Recognise subtitles** (official help terminology). Pick the **spoken language** in your video — it must match what you actually say; the wrong language yields empty or garbage text. Click **Generate** / **Create** and wait for cloud processing (stable connection required). When done, you see text segments aligned to audio on the timeline or on a dedicated captions track.
What you should see: Multiple subtitle segments along the clip, roughly synced to voice.
Steps 3
Review and fix text
Play the full video once with captions visible. Click each **subtitle block** that fails: proper nouns, numbers, jargon. Edit text in the side panel or by double-clicking the block — changes save in real time per web/desktop help. If a block starts late or early, drag its edges on the timeline or use **timing** controls. Do not style yet: content must be correct first. If errors pile up, lower music, keep only the voice track and **regenerate** Auto captions instead of rewriting everything manually.
What you should see: At least three fixes (names, numbers or one badly split phrase); sync acceptable to the ear.
Steps 4
Subtitle style for mobile
With text fixed, open caption **presets**, **templates** or the **style** panel (CapCut offers fonts, color, outline and animations per platform). Pick **high contrast** and large size: on 9:16 the video is watched on a small screen. Avoid thin fonts on bright backgrounds. If your build shows **Auto highlight keywords**, enable it only when it helps scanning — do not decorate every word. Apply the style to **all** blocks on the track («apply to all» or equivalent) so you do not mix three fonts. Play a dense section and confirm captions do not cover your face in the important framing.
What you should see: Uniform captions readable in mobile-sized preview, no stray blocks in a different font.
Steps 5
B-roll: stock or imported AI clip
Find **one moment** in the script where a supporting shot helps (when you say «for example on the street», «the product», «before and after»). Two official paths:
1. **Stock materials**: in import or the library, open **Stock materials**, search by keyword that matches the line and preview short clips. Add one 2–5 s clip to the project bin. 2. **External AI clip**: if you already generated a 3–5 s MP4 in Runway or Kling (see runway-primer-clip / kling-primer-clip), **Import** that file. Inside the editor, some accounts also have **AI media** with text-to-video or image-to-video for a short shot — if you do not see it, your region or plan may not include it; use stock or import.
Do not pull B-roll from sites without a clear license if the video is for a client.
What you should see: One stock clip or an AI MP4 in the media panel, ready to drag to the timeline.
Steps 6
Place B-roll on the timeline
Drag the B-roll clip to a **video track above** the A-roll (overlay) or replace picture on a segment if you prefer a hard cut: at the script point, **split** the A-roll, drop B-roll in the gap and keep **A-roll audio** on the main track so voice continues (classic talking head + B-roll). Trim B-roll length to the phrase it illustrates — 2–4 s is usually enough. Add a **soft transition** if the cut feels harsh; do not overdo it. This tutorial needs only **one** B-roll insert; master one before building eight layers.
What you should see: A segment where video shows stock or AI clip while A-roll voice continues; captions stay visible and readable.
Steps 7
Export and validate
Open **Export** (web or desktop). Pick resolution for your platform (1080×1920 for 9:16 is common). Confirm you export **burned-in subtitles** in the video if the destination is TikTok/Reels/Shorts — some flows also export **SRT** for YouTube; for this tutorial the MP4 with embedded captions is the deliverable. Start export, download the file and play it on a phone or with headphones: voice, caption sync and the B-roll cut. If something fails, return to text fixes or B-roll trim — do not re-export five times unchanged.
What you should see: Final MP4 in downloads, duration similar to trimmed A-roll, captions visible for the full play.
Real use cases
45-second product tip
Auto captions, fix the SKU name, stock of hands using the product at second 20.
Vertical educational clip
Captions with moderate keyword highlight and stock B-roll that shows the concept when you name it.
Hook + demo
First 3 s talking head; imported Kling B-roll when you switch to the visual demo.
Common mistakes
Wrong caption language
You generate with the default language and ASR invents words. Pick spoken language before Generate.
Decorative B-roll
Pretty footage that does not match the line. The cut must illustrate what is said in that second.
Export without listening
A caption block 0.5 s off ruins the short. Play the full file before upload.
Too many text styles
Every block with a different animation tires viewers. One preset for the whole video.
Close
CapCut shortens the vertical short path: import A-roll, Auto captions, fix text, unify style, insert stock B-roll or an AI MP4 where the script needs it and export MP4. AI does the first transcription pass; you guarantee names, sync and that the supporting shot tells the same story as the voice. For audio-only podcast or long transcript editing, Descript is a different flow.
Next steps
- If the AI shot does not exist yet, generate a clip in kling-primer-clip or runway-primer-clip and import it.
- For voiceover from script without camera, see elevenlabs-primer-voiceover and edit over B-roll.
- Tool sheet: /en/tools/capcut/.
Free guide
Want the 15-tools guide?
We send it free. We will also ping you when the next guide in this path goes out.
Unsubscribe whenever you want. We do not sell the list.