Overlays & Captions

Burn text into any finished video: timed title cards (for example a grade card when a presenter names a score), subtitles you write yourself, or captions timed automatically to the speech. The result is a new video; the original is kept unchanged.

You can add overlays in three ways:

Where	How
Jobs (app)	Open a finished video and click Add overlays
Claude	Ask Claude to add overlays or captions (the add_overlays tool). See Use Ads Optimiser in Claude
Pipelines	Add an Add captions (burn in text) step, or run the Re-caption a video template on a video you already have. See the Pipeline editor

Two kinds of text

Timed cues

Each cue is one piece of text with its own timing and look:

Field	Options
Text	Up to 200 characters, on up to 3 lines
Start (s) / End (s)	Seconds from the start of the video. End must be after start
Position	top, center or bottom
Style	caption: large, bold, outlined subtitle text. card: bigger text on a semi-opaque box, for grade cards and titles
Y (%)	Optional. Where the middle of the text sits, from 5% to 95% of the frame height (0 is the top). Overrides Position
Size	Optional. small, medium or large. Captions default to medium and cards to large

A video can have up to 50 cues. Cues can overlap, for example a card in the centre while captions run along the bottom.

Place text exactly

Position puts text in one of three bands. For exact placement, set Y (%) and Size:

• Keep cards off the face. In a 9:16 talking clip the face sits in the upper third of the frame. Put cards at about 60 to 70%, for example Y 62. A centred card on a head-and-shoulders shot can cover the chin
• Break long text into lines. Press Enter in the text box for a second line, up to 3 lines. Maths on one line and A+ on the next reads better than one long line
• Make it smaller. Use small for secondary text, such as a source note near the top
• Narrow the block. Claude and the API can also set max_width, from 0.4 to 1.0 of the frame width, to wrap text into a narrower column

Text placed with Y (%) stays where you put it, even when it overlaps other cues, so check that two cues shown at the same time don't sit on top of each other.

Auto captions from speech

Tick Auto captions from speech and the speech in the clip is transcribed (OpenAI Whisper), then shown as short word groups of one to four words, each on screen while it's spoken.

• Choose the Caption position (bottom by default, clear of TikTok's buttons), or set Caption Y (%) for an exact height and Caption size for smaller or larger captions
• Optionally paste the Script, the words you know are spoken. It fixes the spelling of names, brands and numbers while keeping the speech timing
• Transcription costs about US$0.006 per minute of video (indicative provider cost)

You can use timed cues and auto captions together in one request.

Add overlays in the app

1. Open Jobs and click a finished video
2. Click Add overlays
3. Fill in the first cue row. It starts as a centred card, ready for a title or grade card. Optionally set Y (%) and Size, and press Enter in the text for a second line. Click Add cue for more rows, or Remove to delete one
4. Optionally tick Auto captions from speech, choose the position and paste the script
5. Click Create video with overlays

The hint above the rows tells you roughly how long the clip is, so you can time cues. You're taken to the new job, where you can follow its progress. Its Overlays card lists every cue with its time, style and position, the captions timed to speech, the transcription cost and a link to the Source clip. Open Transcript to read what was recognised.

If something is wrong, for example a cue whose end is before its start, the problems are listed under the form and nothing is started.

Example: grade cards timed to speech

Juno Vale says: "Lots of sitting is linked to worse health. As bad as smoking? No. The link: B. The comparison: D." Two cards appear as she names each grade, with auto captions along the bottom (timings are illustrative):

Text	Start	End	Y (%)	Size	Style
The link: B	4.8	6.2	62	large	card
The comparison: D	6.3	8.0	62	large	card

Each card is on two lines (press Enter after the colon) and sits at 62%, below Juno's face.

Watch the clip first and note the second each point is spoken. See Create an AI influencer for the whole workflow.

Good to know

• Works on any finished video job in your workspace, including lip-synced and captioned clips. You don't need to generate the video again or run a pipeline. Claude can also pass a video URL, and the MCP package a local .mp4 or .mov file
• Source videos can be up to 40 MB
• Every cue must sit inside the clip: a cue that starts or ends after the video ends is refused, with a message giving the clip's length. (In a pipeline's Add captions step, cues past the end are cut off instead)
• The new video is linked to the same character as the source, so it appears in their gallery
• Adding overlays counts as one creative job from your plan, not an image or video generation. Auto captions also incur the small transcription cost above
• If the render fails, Retry re-renders from the stored cues and transcript without transcribing again