No filming:An AI avatar or clone yourself
    Guides

    Pick overlay or auto-captions

    If nobody said the line, it is overlay. Auto-captions transcribe speech. Do not send a mute plate to captions.

    Versely Team••6 min read

    Pick overlay or auto-captions before you send the file to “add text.” Text overlay is copy you authored and burned onto a clip. Auto-captions are a timed transcript of speech. If nobody said the line, captions have nothing to transcribe. Sending a mute product plate to the captioner because you wanted “SALE 30%” on screen is how you get an empty burn, a hallucinated transcript, or a bill for a job that was always overlay.

    Search still asks “what is text overlay” and “auto captions” as if they were one tool. They are two doors on the same picture. This week’s decision is which door the line actually belongs to.

    Nobody said the line: overlay

    A hook you typed. A price. A CTA. A SKU name the presenter never pronounced. An Article 50 paste line. A lower-third that is not speech. Those are overlays. You write the words. The tool places them. They do not have to have been heard.

    Add text overlay to video is that job: one persistent line, or timed overlays when several claims need their own in and out. The line is not a subtitle track. It will not highlight word-by-word unless you built that look as motion graphics. It will not follow a voice, because there is no voice to follow.

    Keep one persistent badge as overlay. Put a sequence of claims on timed overlays. Do not generate the headline into the still unless the type is part of the product photography. Letters a model painted into a pack shot are not editable. Overlay stays a layer until you burn the export.

    Safe area still applies. A hook in the bottom fifth of 9:16 sits under TikTok chrome. Place overlay where a caption preset would also survive: clear of the username stack, clear of the caption you might also burn if there is speech.

    Overlay is the right radio for silent B-roll that needs a claim. Auto-captions on silent B-roll is a 1× (or 2×) spend with nothing to show, or worse, a transcription of room noise. Skip it.

    Someone said the line: auto-captions

    Speech on the file. A talking head. A podcast clip. A screen recording with a voiceover. Those are captions. The AI caption generator transcribes, times, and burns. Add captions to a video is the agent job: 165 language codes, 21 basic looks at 1×, 9 dynamic looks at 2×.

    Word-highlight captions still come from speech. They are not overlay with a bounce. If you wanted karaoke on a line nobody spoke, you wanted timed overlay or a graphic, not auto-captions.

    Captions follow the transcript. Product names will be wrong until you fix them. Read the transcript before you burn. Overlay copy is already the words you meant; the failure mode is placement and contrast, not spelling the SKU as a homophone.

    If the clip has speech and a headline the speaker never said, you owe both radios: captions for the talk, overlay for the claim. That is two steps. It is not “captions with a bigger font on the first line.” The first line of a caption track will be whatever they said first, which is often “hey guys.” Your hook is overlay.

    Mute autoplay does not change this split. A talking clip still needs captions so the mute viewer can read the speech. A mute clip still needs overlay if you authored a claim. The feed being silent is why burn-in exists. It is not a reason to run speech-to-text on a file with no speech.

    The weekly mix-up

    Three briefs that skip the radio:

    1. “Caption this ad.” The ad is a pack shot and a VO that lives in a separate WAV you have not attached. Auto-captions hears nothing. Overlay the claim, or mux the VO first and then caption.
    2. “Put the testimonial on screen.” If the testimonial is spoken in the file, caption it. If it is a quote you pulled from a review, overlay it. A review quote is not a transcript.
    3. “Add the AI label.” Article 50 paste is overlay. It is not a caption cue. Do not hope the transcriber will write “This video is AI-generated” because you mentioned it in the brief.

    Forced narrative — only the lines a same-language viewer still needs, like on-screen alien text — is a caption-shaped object with overlay-shaped authorship. Treat it as authored, timed text, not as auto-captions.

    Do not ask an image or video model to typeset the overlay as part of the generate. You will not restyle it. You will not move it out of chrome. Overlay after picture lock.

    Do not burn captions until picture lock either. Captions are pixels once rendered; a later trim means recaptioning. Overlay on a cut that still moves is the same trap. Lock the cut. Then pick the door. Then place type.

    Caption timing is a speech problem: reading speed, line length, in and out against the voice. Overlay timing is a pacing problem: how long a claim needs to be readable. Do not use caption WPM rules to time a two-word price. Do not use overlay “leave it up the whole clip” on a 40-second monologue unless you want a poster covering the face.

    The Monday radio

    For every line that will appear on screen this week, ask: did someone say this on the file?

    • No. Overlay. Write the copy. Place it. Burn on export.
    • Yes. Auto-captions. Transcribe. Fix the SKU. Pick 1× or 2×. Burn.
    • Both. Captions for speech, overlay for the authored claim. Two passes.

    If the file has no speech and no authored line, you do not owe type. A pretty silent clip with nothing to read is allowed when the picture is the hook. It is not allowed when the hook was a sentence you forgot to overlay.

    When the radio is overlay, skip the caption preset picker. When the radio is captions, skip the overlay composer except for the extra claim. The safe area is shared. The tools are not.

    If you need both doors on the same cut, sign up and say whether the line was spoken. The captioner is a transcriber. It is not a copywriter.

    FAQ

    The VO is on a separate track. Caption or overlay?

    Mux or replace audio so the speech is on the file, then auto-caption. Overlay cannot time itself to a WAV sitting in another folder. If you will never mux, overlay the claims by hand and accept they will not karaoke the VO.

    Can auto-captions pick up text that is already in the picture?

    No. They transcribe audio. On-screen packaging is not speech. If you need that word as a caption, someone has to say it, or you overlay it.

    Should the CTA be a caption cue at the end?

    Only if they said the CTA. If you wrote “Shop the link,” that is overlay, timed to the last seconds, sitting in the safe area — not a subtitle of a sentence that never happened.