A pipeline that turns a topic into a captioned vertical video is a known recipe: a model writes the script, text-to-speech reads it, stock clips fill the scenes and ffmpeg burns in word-timed captions. An agent can have that running in a weekend for pennies per video. What Revid sells on top is harder to copy: video-model clips from several vendors under one balance, a library of millions of viral videos to remix, voice cloning and avatars, and one-click posting to TikTok and Instagram, whose publishing APIs need app approval.
Before you buildA free Revid account gives a few trial credits, so you can see what a finished video looks like before you build or pay.
Build me a short-video maker to replace Revid. Requirements: - Node.js 22 + TypeScript: a command line and one small web page. `shorts make "<topic or URL>"` writes a 1080x1920 MP4; `npm start` serves http://localhost:4880 to review the script and re-render. - Script: an OpenAI-compatible LLM (BASE_URL, API_KEY, MODEL in .env) writes a hook, 5 to 8 scenes and one visual keyword per scene as JSON checked with `zod`. A URL input is fetched and reduced to its text first (`@mozilla/readability`). - Voice: OpenAI text-to-speech by default, with a local `piper` option; word timestamps from the OpenAI transcription API at word granularity. - Visuals: one stock clip per scene from the Pexels video API (PEXELS_API_KEY), cropped to vertical and trimmed or looped to the scene length; cache downloads in ./cache. - Compose with `ffmpeg` (a required install): scene cuts, the voiceover, optional music from ./music at -20 dB under speech, and burned-in word-by-word captions as an ASS file (libass), large and centered inside the 880x1220 safe zone. - Storage: ~/Shorts/YYYY-MM-DD-slug/ with script.json, captions.ass, audio.mp3 and video.mp4, plus better-sqlite3 in ./data/shorts.db for run history and the cost of each run. - Show an estimated API cost before rendering and refuse above MAX_COST_USD in .env. - Optional upload to YouTube as a Short through the Data API v3 (OAuth, `googleapis`). - Single user. No accounts, no telemetry. Outbound calls: the LLM, TTS, Pexels and YouTube only. - Out of scope: AI-generated video clips, avatars, face swap, TikTok and Instagram publishing, a viral-video library. - README: installing ffmpeg, the .env keys, and a warning that uploads from a YouTube API project that has not passed Google's audit stay private; budget an hour for the OAuth consent screen.
Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.
The first prompt builds the core. Run these one at a time in the same folder to grow it into your own Revid AI.
Add generated clips to the short-video maker in this folder. Requirements: - A scene field `visual: "stock" | "ai"`; the LLM marks at most 2 scenes per video as "ai" with a prompt. - Call fal.ai (FAL_KEY in .env) with a configurable model id from .env, poll until done, download the clip into ./cache. - Show the estimated cost per clip before rendering and respect MAX_COST_USD. - If a call fails or times out, fall back to a stock clip and note it in script.json. - Update the README with a warning that per-clip prices change by model.
Add batch mode to the short-video maker in this folder. Requirements: - `shorts batch topics.txt` reads one topic or URL per line and renders them one after another. - Skip topics already rendered (match on a hash of the topic), write a batch-report.md with cost and duration per video, and stop when the total passes MAX_COST_USD. - A `--schedule` flag spaces YouTube uploads 24 hours apart as scheduled publish times. - Update the README.
Add caption presets to the short-video maker in this folder. Requirements: - Three ASS presets in ./styles: classic white with outline, one-word-at-a-time bold, and a highlighted current word; choose with --style or a default in .env. - Keep the text inside the 880x1220 safe zone and wrap at 18 characters per line. - Add a `shorts preview-captions` command that renders a 5-second test clip per preset. - Update the README with how to add a preset.
Founders and small business owners want a finished, postable video without learning an editor. Revid bundles the script, voice, visuals, captions and posting under one bill, and its ideas library shows what formats are already working, which a home-built pipeline does not.
Estimate · Per video the script, voice and transcription cost a few cents; stock footage from Pexels is free. Rendering runs on your machine. The build is token usage on your agent plan.
| Plan | Monthly | Annual, per month | What you get |
|---|---|---|---|
| Free account | $0 | $0 | Free trial credits for video generation and search of the viral-video library; no regular creation. |
| Growth | $39 | $32 | 2,000 AI credits a month, publishing to TikTok, Instagram and YouTube, 3 Auto-Mode workers, 5 watched channels, avatars and voiceovers in 70+ languages, API, MCP and CLI access. |
| Ultra | $199 | $166 | 12,000 AI credits a month, 10 Auto-Mode workers, voice creation and cloning, 50 watched channels. |
| Ultra, higher credit tiers | $459 | — | A slider on the Ultra card: 30,000 credits for $459, 50,000 for $749, 100,000 for $1,459 and 200,000 for $2,799 a month. |
No free plan. Sign-up needs no card and gives free trial credits and viral-video search.
Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: AI video clips from Seedance, Veo and MiniMax, plus image-to-video, The library of millions of viral short videos with engagement data to remix.
About $39 a month on the Growth plan, $468 a year (checked 2026-10-19).
AI video clips from Seedance, Veo and MiniMax, plus image-to-video; The library of millions of viral short videos with engagement data to remix; One-click posting to TikTok and Instagram; Avatars, face swaps and cloned voices; Auto-Mode workers that make and publish videos on a schedule. If any of those matter to you, keep paying.
Yes: MoneyPrinterTurbo (open source), Remotion (free), AI-Youtube-Shorts-Generator (open source).