AI video · checked 2026-10-19

Can I vibecode Captions?

KINDA· weekend project

Split the product in two. The editing core is buildable: transcribe the recording, burn in word-highlighted captions, cut filler words and pauses, clean the audio and drop in stock B-roll chosen by an LLM. An agent does that in a weekend with Whisper and FFmpeg. The paid Max plan, though, is mostly generative AI: avatars, AI twins, eye-contact correction and generated footage run on proprietary models and GPUs that a personal project can't match without paying per call. Build this if you pay for captions and trimming, and keep the subscription if you pay for the avatars.

$24.99/moprice

$300you'd save a year

weekendbuild time

—people replaced it

Before you start

What you need on hand.

DifficultyIntermediate · Some system setup or keys.

TimeAn afternoon for captions and cuts; a weekend for the text editor and B-roll

Claude Code, Codex or CursorAny of them can run the prompt.
Python 3.12faster-whisper downloads a model on first run, so allow a few GB of disk.
FFmpegInstall it with your package manager; the prompt makes the tool check for it first.
An LLM key and a Pexels API keyOnly for the optional B-roll step.

Before you buildCaptions' Free plan keeps basic editing and one caption template at no cost. If you only trim and caption, it may be enough before you pay or build.

1. Install one agent: Claude Code, Codex, Cursor.
2. Make an empty folder and open it in the agent.
3. Paste the prompt below, press enter, and answer its questions as it builds.

The prompt

Build me a local talking-head video polisher to replace Captions. Requirements:

- Python 3.12 CLI: `polish <video> --style bold --format 9:16` writes ./out/<name>.mp4 and
./out/<name>.srt. FFmpeg runs as an external program; fail early when it is missing.
- Transcribe with `faster-whisper` (WHISPER_MODEL in .env, default small), keep word timestamps
in ./out/<name>.words.json and detect the spoken language automatically.
- Captions: an ASS file via `pysubs2`, three or four words per line with the spoken word
highlighted. Five styles in ./styles/*.json (font, size, colors, outline, position, all-caps
flag), chosen with --style, burned in by FFmpeg. Also write the plain SRT.
- Tighten: `--cut-fillers` removes um, uh, er and stuttered repeats, and `--trim-pauses` cuts
silences over 0.6 seconds, both from word timestamps, with a 30 ms audio crossfade per cut.
- Audio: `--clean-audio` applies FFmpeg `afftdn` noise reduction and `loudnorm` to -16 LUFS.
- Edit as text: `polish serve` opens http://localhost:4840 with the transcript beside a player.
Clicking a word deletes it from the video, undo restores it, and Export renders the result.
- Optional B-roll: with PEXELS_API_KEY set, the LLM from .env picks up to five moments and a
search phrase each; fetch a portrait clip from Pexels and overlay it for 2 to 4 seconds.
- Single user. No accounts, no telemetry, local-first: only the LLM and Pexels requests leave
the machine, and the README says exactly what each one sends.
- Out of scope: AI avatars and digital twins, eye-contact correction, generative video or music,
dubbing and translation, a mobile recorder, a teleprompter.
- README: installing FFmpeg, render time on CPU, and a reminder to read Pexels' license terms
before posting.

Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.

Share on X

Max MiskowSponsoredAI agents, automations and MVPs, built for your company.Visit site ›

Then keep going

Next prompts, same folder.

The first prompt builds the core. Run these one at a time in the same folder to grow it into your own Captions.

Extend the video polisher in this folder with brand styles. Requirements:

- A ./brand.json file with font file path, primary and highlight colors, outline width, caption position and an optional logo PNG with corner and opacity.
- `--style brand` uses it; the logo is overlaid for the whole video with FFmpeg.
- Validate the file on start and print the exact field that is wrong.
- Add a preview command that renders the first five seconds only.
- Update the README with a sample brand.json.

Add chapters to the video polisher in this folder. Requirements:

- Ask the LLM for chapter titles with start times from the transcript; write them as FFmpeg metadata chapters in the MP4 and as a ./out/<name>.chapters.txt in YouTube's 00:00 format.
- Optionally render a two-second title card per chapter with the title in the chosen caption style (--chapter-cards).
- Never move a chapter off a sentence boundary.
- Update the README.

Add batch mode to the video polisher in this folder. Requirements:

- `polish batch ~/Recordings --style bold` processes every video not yet in ./out/processed.json, one at a time, with the same flags as the single-file command.
- Write a summary table at the end: file, duration, cuts made, time saved, errors.
- Continue after a failure and exit non-zero if any file failed.
- Update the README with a cron or launchd example.

What you lose

AI avatars and digital twins, and generative video from a text prompt
AI eye-contact correction
The phone app with a camera, teleprompter and script generator
Dubbing and translated subtitles in 100+ languages
AI Edit that chooses styles, cuts scenes and adds music in one step

Why people still pay

People pay for the whole loop on a phone: record, get captions and cuts applied, and post in minutes. The AI features are the draw, since eye contact, twins and generated footage need models and GPUs behind them. Reviews are split: the convenience is praised, while credit limits and billing draw the complaints.

Moatproprietary models execution polish infrastructure scale

Do the math

Captions or your own copy?

Over

Your usage

Count my time12h ×$/h

Captions, 1 yr$300

Your own copy, 1 yr$26running + build session

You keep$274pays for itself in the first month

Estimate · Runs on your machine; Whisper and FFmpeg cost time, not money. The optional B-roll step costs a few cents of LLM tokens per video, and Pexels is free. The build is token usage on your agent plan.

Leaving Captions

Take your data, then cancel.

1 · Take your data out of Captions

1Open the project, select Export, pick the settings and select Export again to save the video to your device.
2Under Advanced options you can set resolution, frame rate and bitrate; captions can be downloaded as a separate SRT file.
3Projects that use AI features such as AI Edit, caption styles, eye contact or background music need a paid subscription to export.
4The help page doesn't describe a project backup or full project download, so export every video you want to keep as a file.

2 · Cancel Captions

1Open Settings, then Account, then Billing, then Manage Plan, and select Cancel Subscription.
2Subscriptions renew monthly; cancelling stops future charges.
3If you subscribed through Apple or Google, cancel in that store: the help pages we read don't cover the difference.
4If you are asked to pay again after subscribing, you may have two accounts: restore the subscription instead.

Export every project that uses AI features before you cancel, since those need a paid plan to export.
Monthly plans are non-refundable; if technical problems stopped you from using the app, contact support.
Unused credits roll over for up to two extra months, but nothing in what we read says they survive cancelling, so use them first.
Teams and Business plans are no longer sold; existing customers on them can still add seats.

Checked 2026-10-19 against Captions's docs: 1, 2, 3, 4

The escape hatch

Don't feel like building it? These are already free.

OpenShortsOpen sourceMIT clip generator with face tracking, subtitles and dubbing; free to self-host with your own keys, hosted plans from $12 a month.★ 6.8k · active Oct 2026 · docker to self-host
WebVisit ↗BuzzOpen sourceMIT desktop app that transcribes and translates audio offline with Whisper, and exports SRT subtitles.★ 22k · active Oct 2026 · one-click install
macOS, Windows, LinuxVisit ↗Subtitle EditOpen sourceMIT subtitle editor with automatic transcription and timing tools; for captions, not full video editing.★ 15k · active Oct 2026 · one-click install
WindowsVisit ↗

All 4 free alternatives to Captions ›

The numbers

Captions pricing

Plan	Monthly	Annual, per month	What you get
Free	$0	$0	Basic tools such as trim and transitions, one caption template, 60 to 200 lifetime credits.
Basic	$9.99	—	200 credits a month, Script Generator, Camera and Teleprompter, dubbing.
Max	$24.99	—	500 credits a month, 2 seats. Generative AI, AI Edit styles, Chat to Edit, Clips, AI Twins and more AI models.
Frontier	$69.99	—	1,400 credits a month, 3 seats, top-tier generative AI models, faster generations.
Frontier 2x	$139.99	—	2,800 credits a month, 4 seats; the same as Frontier with twice the usage.
Frontier 4x	$279.99	—	5,600 credits a month, 5 seats; the same as Frontier with four times the usage.

Free tier: No AI Edit, generation or AI twins on Free; most AI features need a subscription to export.
Billing: Monthly or yearly; yearly bills 12 months upfront at a discount, though no annual price is listed. Upgrades take effect at once and are prorated, downgrades at the end of the cycle. Unused credits roll over for up to two extra months, so a balance can reach three times the monthly allowance. Monthly plans are non-refundable.
Hidden costs: AI Edit, generation and the other AI features all draw on one credit pool, and reviewers say 500 credits can run out faster than expected. Anything above Max is a steep jump, from $24.99 to $69.99.
Source: www.captions.ai

Also on the list

BlitzReels$39/mo · Audio & videoKINDA

CapCut$19.99/mo · Audio & videoKINDA

Descript$24/mo · Audio & videoKINDA

Questions

Captions, briefly

Not answered here? Tell us and we'll add it.

Ask a question ›

Free covers basic editing tools and one caption template, with 60 to 200 lifetime credits. AI Edit, video generation and the other AI features need a paid plan.

Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: AI avatars and digital twins, and generative video from a text prompt, AI eye-contact correction.

About $24.99 a month on the Max plan, $299.88 a year (checked 2026-10-19).

AI avatars and digital twins, and generative video from a text prompt; AI eye-contact correction; The phone app with a camera, teleprompter and script generator; Dubbing and translated subtitles in 100+ languages; AI Edit that chooses styles, cuts scenes and adds music in one step. If any of those matter to you, keep paying.