Should I vibe code
Create captioned audiograms and short social clips from user-owned audio
It's ffmpeg with a template. The product is the hundred boring decisions after the first render looks great.
?
Their verdict, the Basic price and the build-time estimate come from their entry, MIT-licensed. Checked 2026-08-04.
?
Our verdict, the regret score and everything below it. Editorial and unsponsored — nobody can pay to be moved.
The honest answer
why the verdict is what it is
This is ffmpeg with a template, and an agent will hand you a captioned waveform video in an afternoon. The distance between that and Headliner is not cleverness, it is a hundred boring decisions: caption timing that holds past the twenty-minute mark, fonts and emoji that render the same on your laptop and on a headless render box, safe areas for four aspect ratios that platforms change without telling anyone, and a job queue, because a ninety-minute episode is not a web request. None of it can hurt a person. Worst case you publish a clip with a guest's name misspelt into the pixels and re-render. Which is exactly why it makes a delightful v0 and a tedious product.
What actually breaks
not "if". the specific failures.
- Caption sync, which drifts on long files in a way nobody notices until somebody watches past minute twenty
- Speech recognition on names, brands and jargon — and a burned-in caption can only be fixed by re-rendering and re-uploading
- Fonts and emoji, which look one way on your machine and another on a headless Linux box with a different fontconfig
- Aspect ratios and safe areas, four of them, changing whenever a platform redesigns its player
- The render queue, the first time two people export a ninety-minute episode at once and your web process dies holding both
- Loudness, when a clip lands six LUFS hotter than the feed it was cut from
- Music beds, the one part of this with a licence attached and a rights-holder who scans for it
Is that you?
the verdict is a default, not a law
- It renders locally, from your own audio, for your own show
- The edit is stored as a manifest, so any clip can be re-rendered after you fix the template
- Captions get a human read-through before anything is burned in
- You are batching hundreds of clips a month, which is where a script genuinely beats a subscription
- You are hosting uploads for other people, because now you own a transcoding farm and everything anyone uploads to it
- There is no queue and rendering happens inside a request handler
- Captions go from the model straight into the pixels with nobody reading them
- You are adding music you have not licensed to clips you intend to publish
If you build it anyway
the checklist, then the prompt that enforces it
- Render out of process, through a real job queue, from the very first version. Video encoding inside a request handler is a mistake you get to make exactly once.
- Put a human review step between transcription and burn-in. ASR is excellent overall and reliably wrong on precisely the words that matter — names, products, places.
- Pin the ffmpeg version, the fonts and the container, and render locally in the same image you render in production, or your preview is lying to you.
- Keep the edit as data: a JSON manifest of source file, in and out points, caption text with timings, template and output spec. Renders become reproducible and the back catalogue stays editable.
- Normalise loudness to the target platform rather than trusting the source, and check the result rather than assuming the filter did what you meant.
- Delete source uploads on a schedule and store outputs where you can see the bill. Video is where hobby projects meet real invoices.
- Burn captions from the same timing data you display in the editor, and re-verify sync at the end of the file, not the start — drift is a tail problem.
I am building an audiogram and clip generator: audio in, captioned video out.
Apply these constraints and push back when I ask for shortcuts.
1. All rendering happens in a background worker behind a job queue, with status,
retries and a hard timeout. Never render inside an HTTP request, even for the
first prototype — it is the thing that will be load-bearing later.
2. Use ffmpeg for encoding and a real layout library for the visuals. Do not
hand-roll a video encoder or a waveform renderer that writes frames one at a
time without telling me what that costs per minute of output.
3. Store the edit as a JSON manifest — source path and hash, in/out points,
caption cues with timings, template id, output spec — and make the renderer a
pure function of that manifest. I want to re-render last year's clips after a
template change.
4. Captions are a two-step flow: transcribe, then a review screen, then burn in.
Never write model output straight into pixels. Say plainly that burned-in text
cannot be corrected after publishing, only replaced.
5. Verify caption sync at the end of the file as well as the start, and fail the
render if drift exceeds a threshold I set.
6. Pin ffmpeg, fonts and the base image. Render locally in the same container as
production and tell me when a font is missing rather than substituting one.
7. Normalise output loudness to a target I choose and measure the result.
8. Delete or expire source uploads on a schedule, and show me the projected
storage and egress cost per hundred clips before I choose defaults.
9. If I ask for background music, ask me where the licence came from.
10. Out of scope unless I ask: multi-user accounts, a hosted upload service,
scheduled posting to social platforms, and anything that stores other
people's audio. Each turns a local tool into a service with a bill.
That one keeps you out of trouble. For the prompt that actually builds it, canivibecodeit.com has one.
their build prompt ↗Or don’t build it
the boring option, and the way back out
$9.99 a month for Basic — ten unwatermarked audiograms — or $25.99 for unlimited Pro. If you ship one episode a week, the subscription costs less than the Sunday you will spend on font metrics and safe areas. Build it if you want a house style no template gives you, or if you are cutting hundreds of clips a month, which is the volume where a two-hundred-line ffmpeg script starts to win.
$9.99/mo is cheaper than your weekend.
The source audio plus the JSON edit manifest is everything worth keeping. Rendered mp4s are disposable: as long as the manifests survive, a new pipeline can reproduce every clip you ever posted, and a template change becomes a batch job rather than a decision to abandon the back catalogue. Without them, your archive is a folder of files you can play and never edit.
Programmatic video rendering in React, with the render-farm and composition problems already solved.
Local speech-to-text, good enough for caption cues and cheap enough to run per episode.
Questions
Is ffmpeg genuinely enough for the render?
For the encode, yes, and the waveform filter will get you a recognisable audiogram in one command. The work is everywhere else: composing the frame, laying out captions that wrap sensibly, keeping brand assets consistent, and running the whole thing somewhere that will not fall over when two exports collide.
How good are automatic captions in 2026?
Very good on general speech and still unreliable on the words your audience notices: guest names, product names, acronyms and anything domain-specific. Since the caption ends up in the pixels, a wrong word costs a re-render and a re-upload, which is why the review step is a guardrail and not a nicety.
Why DEMO ONLY rather than SHIP IT, if nothing here is dangerous?
Because the gap is durability rather than danger. The first render is genuinely delightful and the next fifty are a maintenance relationship with ffmpeg, fontconfig, four aspect ratios and a queue. Most people abandon it in month three and go back to the subscription, which is the definition of a good v0 and a bad product.
Every week, someone ships something they shouldn’t have.
New verdicts, the worst thing that landed in the trap, and the occasional incident report. No other email, ever.
Cleaning audio is scriptable. Publishing a feed on a schedule is where it becomes a commitment.
Loudness normalisation is a well-specified standard and ffmpeg already implements it.
Transcript to clips and quotes is a prompt chain over Whisper output. Very reachable.
last reviewed 2026-08-05 · verdict is editorial and unsponsored · shared entry data from canivibecodeit under MIT · not legal advice