No filming:An AI avatar or clone yourself
    Versely
    Guides

    Get a transcript, then make the SRT

    Transcribe with Cartesia ink-whisper, correct the page, then build SRT cues. A .txt is not a caption file.

    Versely Team••6 min read

    Get a transcript first. Then make the SRT. Get a video transcript (transcribe_audio, Cartesia ink-whisper) returns editable text. It does not return numbered cues with start and end times. A .txt renamed to .srt will fail a player or dump the whole memo as one cue. Correct names and SKUs on the page. Then author cues — or burn-in if the destination is mute short-form. This is the document-then-sidecar order. It is not a transcript is not an SRT restacked as a definition page.

    Transcribe when you need words you can fix

    transcribe_audio wants an audio_url — pulled from the video, or a wav. Pass language if you know it. Regenerating because you forgot the language field is a second bill. Accuracy follows the mix. Overlap, beds, and room slap degrade the page. If you needed a clean document from a smash, isolate or stem first.

    Good reasons to transcribe before any caption file:

    • You will correct product names, prices, and homophones in a doc the client can comment.
    • You will translate, then dub, then caption the new language.
    • You will paste quotes on a landing page.
    • You will build an SRT for YouTube long-form where the player should toggle type, not weld it.

    Procedure:

    1. Point get a video transcript at the audio you will actually ship — the mix after a bed, after a dub, after lipsync. Transcribing the scratch take is a document of the wrong line.
    2. Set language.
    3. Run ink-whisper. Read it. It will be wrong on SKUs. That is expected.
    4. Save the corrected document next to the video. This is still not an SRT.
    5. Stop if a document was the whole brief.

    What this pass will not give you:

    • In and out points per line.
    • Speaker labels you can swear in court.
    • A burn-in. Pixels are a different tool.

    Billed per audio clip, based on length. Ask for estimate_cost on a long file. For a short English clip, the free video to text converter does the transcript step in your browser, up to 10 minutes, with no upload.

    Versely’s short-form caption tools are burn-in tools. Styled auto-captions are composited into frames. A sidecar workflow does not start there. It starts here, with the page.

    An SRT is cues, not a memo

    An SRT is numbered blocks: index, HH:MM:SS,mmm --> HH:MM:SS,mmm, then one or two lines of text, then a blank line. Times are the object. Words without times are a script.

    How to get from a corrected transcript to an SRT without lying:

    1. Split the page into caption-sized lines — short, readable on a phone if this will also be a burn later; a bit longer if it is YouTube under the player.
    2. Time each line against the locked audio. Scrub. Do not guess from reading speed if the speaker pauses. Caption timing is the craft; karaoke is per-word, SRT is usually per-phrase.
    3. Write the file as real cues. Do not paste the whole transcript into cue 1.
    4. Validate in a player against the same file you will upload. A crop or a trim after timing desyncs the sidecar. Picture lock first. Aspect first. Then transcribe. Then SRT.

    YouTube can auto-sync a plain transcript on upload. That is YouTube’s tool, not an SRT you control. If you needed exact in/out (legal line on screen for a required duration, a skipped filler word), author the sidecar.

    Free tools: VTT to SRT converts a sidecar you already have. The slug is vtt-to-srt-converter, not vtt-to-srt. It will not invent timestamps from a novel.

    Burn-in versus sidecar:

    • Burn-in for TikTok, Reels, Shorts — mute autoplay. Add captions to video: 165 language codes, 21 BASIC at 1×, 9 DYNAMIC at 2×. You can still transcribe first, correct, then burn from the approved words if the tool path allows a corrected source; do not trust a live recognizer on a dirty mix for a legal SKU.
    • SRT/VTT for YouTube long-form, courses, accessibility on a player with a CC button.

    Do not ship both a burn and an SRT of different text. The deaf viewer and the hearing viewer should get the same offer.

    Order with the rest of the finish

    Wrong order:

    1. Burn karaoke on a scratch mix.
    2. Transcribe for the blog.
    3. Export an SRT by renaming the blog paste.
    4. Crop 9:16 to 16:9. Wonder why cues drift and type is in the picture twice.

    Right order:

    1. Picture lock, aspect lock, mix lock (bed ducked, no second singer).
    2. Transcript. Correct.
    3. Either burn those words into the short-form file, or time them into an SRT for the long-form file, or both from the same page.
    4. Do not recaption by cropping a burned file.

    YouTube labels (27 May 2026): photoreal Shorts overlay is on the video. An SRT does not disclose AI. Neither does a transcript. Disclosure is a separate label.

    EU Article 50 from 2 August 2026 is the same: the sidecar is not the AI mark.

    Page, then cues, then stop

    The test: open the .srt in a text editor. If you do not see arrows and timestamps, it is not an SRT. If cue 1 is 400 words, it is not captions.

    Cue length: one or two lines, not a paragraph. A legal disclaimer that must stay on screen for three seconds is a cue with a 3-second window, not a footnote in cue 1. If the player wraps a long cue over the pack, split it. Timing still comes from the audio you corrected on the transcript page — do not re-guess the words while you time. The page is the script. The SRT is the clock.

    If you already burned captions for Reels and now YouTube wants a sidecar, do not scrape the burned pixels with OCR. Use the same corrected transcript. Time it against the 16:9 file. Two deliveries, one page of words. That is the point of transcribing before either burn or SRT.

    If you need a seat to transcribe the keeper and then caption or sidecar from the corrected page, sign up. Bring the locked mix.

    FAQ

    Does get-a-video-transcript return an SRT?

    No. It returns editable text from Cartesia ink-whisper. An SRT is numbered cues with start and end times. You build that after you correct the page — or you burn-in for short-form instead.

    Can I rename transcript.txt to captions.srt?

    No. Players expect cue structure. A renamed memo is one illegal block or a failed upload.

    Should I transcribe before or after I add a music bed?

    After the mix you will ship, if the bed could confuse the recognizer. If the bed is a quiet instrumental, either order can work; a singing bed will steal words. Stem first if the song talks.