Turn any viral video into your next winning ad — in seconds.
Paste a Reel, Short, TikTok, or Facebook Ad → get an on-brand script, hook, caption, and visual direction powered by Gemini 2.5 Flash.
Stop reverse-engineering virality — let AI recast what already works into ads that sound like you.
|
🔗 Multi-Platform Ingest Paste any Reel, Short, TikTok, or FB Ad URL — one input, four sources. 🎧 Audio-Only or Full Video Toggle between lightning-fast transcript mode and deep multimodal analysis. 🛍️ Product URL Scraping Drop a product link — brand context auto-extracts in the background. |
⚡ Gemini 2.5 Flash Core State-of-the-art multimodal reasoning over video, audio, and brand data. 🌐 Native Language Preservation Hindi stays Hindi. Tamil stays Tamil. No forced English translation. 🎬 Scene-Aware Direction Understands pacing, hooks, and structure — not just words on screen. |
📋 One-Click Copy Every card, every output — copy-to-clipboard on tap. 🌓 System-Aware Dark Mode Cal.com-inspired monochrome that respects your OS theme. 🎥 Inline Source Preview Watch the original video right next to your generated assets. |
Five production-ready assets, generated from a single URL.
| # | Output | What It Delivers |
|---|---|---|
| 1 | 📊 Video Summary | Hook breakdown + format + tone + narrative structure — the anatomy of why it works. |
| 2 | 🎯 Adapted Hook | A rewritten sub-10-second opener, retuned for your brand voice. |
| 3 | 📝 Adapted Script | A 30–45s script with inline [VISUAL CUE] markers, ready to shoot. |
| 4 | 📱 Instagram Caption | 100–150 words + hashtag set + CTA — drop-in ready. |
| 5 | 🎬 Visual Direction | 2–3 scenes mapped with camera angles, overlays, and mood. |
Note
🌐 Original language is always preserved. If the source reel is in Hindi, your script comes back in Hindi. Tamil stays Tamil. Hinglish stays Hinglish. No tone is lost in translation.
| 👤 Persona | 🎬 Scenario | 🚀 Outcome |
|---|---|---|
| Solo Founder | Spotted a viral Reel at 11pm — needs an on-brand version live by tomorrow. | $0 spent on copywriters. Ship ads at the speed of inspiration. |
| D2C Marketing Team | Testing 20 ad creatives a week across IG, TikTok, and YouTube Shorts. | Brand at scale. Consistent voice across every platform, every variant. |
| Social Media Agency | Managing 15 client accounts, each with a unique tone and product line. | 10x faster swipe-file building. Turn competitor scans into client decks in minutes. |
| Content Creator | Wants to remix trending formats without copying — needs a fresh angle fast. | Trend-native, never derivative. Hook templates retuned for your niche. |
Get Recast running locally in under 5 minutes.
You'll need Node.js 18+, yt-dlp, and ffmpeg installed on your system.
🍎 macOS
# Install Node.js (if not already installed)
$ brew install node
# Install yt-dlp and ffmpeg
$ brew install yt-dlp ffmpeg🐧 Linux (Ubuntu / Debian)
# Install Node.js 18+
$ curl -fsSL https://deb.nodesource.com/setup_18.x | sudo -E bash -
$ sudo apt-get install -y nodejs
# Install ffmpeg
$ sudo apt-get install -y ffmpeg
# Install yt-dlp via pip
$ pip install yt-dlp🪟 Windows
# Install Node.js from https://nodejs.org (LTS recommended)
# Install ffmpeg and yt-dlp via winget
> winget install Gyan.FFmpeg
> winget install yt-dlp.yt-dlp
# Or via Chocolatey
> choco install ffmpeg yt-dlpYou'll also need a Google API key — grab one from Google AI Studio.
$ git clone https://github.com/thegauravmahto/recast.git
$ cd recast
$ npm installCreate a .env file in the project root and add your API key:
$ echo "GOOGLE_API_KEY=your_api_key_here" > .env
$ npm run devOpen http://localhost:3000 — you're live! 🎉
Configure Recast by setting these in your .env file:
| Variable | Required | Default | Description |
|---|---|---|---|
GOOGLE_API_KEY |
✅ Yes | — | Your Gemini API key used for script generation via Gemini 2.5 Flash. |
YTDLP_PATH |
⚪ No | yt-dlp (from $PATH) |
Absolute path to the yt-dlp binary. Override if it's not on your system PATH. |
FFMPEG_PATH |
⚪ No | ffmpeg (from $PATH) |
Absolute path to the ffmpeg binary. Override if it's not on your system PATH. |
Tip
Don't have a Gemini API key yet? Generate one for free at aistudio.google.com/apikey — it takes about 30 seconds.
For the impatient — clone, install, and launch in a single shot:
git clone https://github.com/thegauravmahto/recast.git && \
cd recast && \
npm install && \
echo "GOOGLE_API_KEY=your_api_key_here" > .env && \
npm run devImportant
Replace your_api_key_here with your actual key from Google AI Studio before running, or edit .env afterwards and restart the dev server.
flowchart LR
A[👤 User Pastes<br/>Video URL] --> B[📥 yt-dlp<br/>Downloads Video]
B -->|standard mode| D[☁️ Gemini Files API<br/>Upload & Poll]
B -->|audio mode| C[🎵 ffmpeg<br/>Extract Audio]
C --> D
D -->|file ready| E[🧠 Gemini 2.5 Flash<br/>System Prompt Analysis]
E -->|markdown response| F[📝 Markdown Parser<br/>Split by Headers]
F -->|5 sections| G[📦 JSON Payload]
G --> H[💎 React UI<br/>Result Cards + Copy]
style E fill:#8b5cf6,stroke:#6d28d9,stroke-width:3px,color:#fff
style A fill:#0ea5e9,stroke:#0369a1,color:#fff
style H fill:#10b981,stroke:#047857,color:#fff
- Paste & Submit — User drops a video URL into the dashboard input and hits analyze.
- Download —
yt-dlppulls the video locally into a temp directory on the server. - Optional Audio Extract — If audio mode is enabled,
ffmpegstrips the audio track to shrink the upload payload. - Upload to Gemini — The file is pushed to the Gemini Files API and polled until the processing state turns
ACTIVE. - AI Analysis — Gemini 2.5 Flash runs with a specialized system prompt and returns structured markdown with 5 canonical headers.
- Parse & Render — The markdown is split into a JSON object keyed by section and rendered as copy-ready cards in the React UI.
| Tool | Purpose |
|---|---|
 |
App Router, server actions, route handlers |
 |
Component model for the dashboard UI |
 |
End-to-end type safety across routes & components |
 |
Utility-first styling with custom design tokens |
 |
Accessible, copy-paste primitives (cards, buttons, inputs) |
 |
Class-based theme switching with no flash |
| Tool | Purpose |
|---|---|
 |
Multimodal LLM that watches the video and writes the analysis |
 |
Handles large media uploads and processing state |
 |
/api/analyze orchestrates the full pipeline server-side |
| Tool | Purpose |
|---|---|
 |
Reliable video downloader across hundreds of sources |
 |
Audio extraction for the lightweight "audio mode" path |
 |
Package management |
Recast is a small, focused codebase — here's the lay of the land:
recast/
├── app/
│ ├── layout.tsx # Root layout: theme provider, fonts, metadata
│ ├── page.tsx # Home route — renders the <Dashboard /> client
│ ├── globals.css # Tailwind layers, design tokens, Cal.com-style shadows
│ └── api/
│ ├── analyze/
│ │ └── route.ts # POST: yt-dlp → (ffmpeg) → Gemini upload → parse → JSON
│ └── video/
│ └── [filename]/
│ └── route.ts # GET: streams downloaded videos back to the client
├── components/
│ ├── dashboard.tsx # Main UI — URL input, loading skeletons, results grid
│ ├── result-card.tsx # Reusable card with title, markdown body, copy-to-clipboard
│ ├── theme-provider.tsx # Thin next-themes wrapper for the App Router
│ └── theme-toggle.tsx # Sun/moon button bound to the theme provider
├── DESIGN.md # Design system reference — tokens, spacing, motion
└── .env.example # GOOGLE_API_KEY and other env var templates
Architecture philosophy: the /api/analyze route does all the heavy lifting — downloading, transcoding, uploading, polling, and parsing — so the client stays a thin, stateless rendering layer. Components are small and composable (<ResultCard /> knows nothing about Gemini), and the entire flow is request/response — no websockets, no queues, no state machines to babysit.
Reels & video posts |
YouTube Shorts & long-form |
TikTok Public videos |
Ads Library & videos |
Tip
1000+ more sites supported via yt-dlp — Twitter/X, LinkedIn, Vimeo, Twitch, Reddit, and far beyond. See the full list of supported sites.
Input
Product Context: A 7-day Notion productivity course for solo founders.
Promises 10 hours/week back. Price: $49.
Source Video: https://www.instagram.com/reel/EXAMPLE/
(a fitness coach explaining how she structures her training week)Output — Recast adapts the source video's structure, pacing, and hook style to your product:
🧠 1. Video Summary
A 38-second Reel where a fitness coach breaks down her "non-negotiable 3-day split." The hook calls out a common mistake (training every day), then walks through her weekly system with quick B-roll of workouts. Closes with a soft CTA to her training program. Tone is confident, direct, no-fluff.
🎯 2. Adapted Hook
"Stop 'hustling' 7 days a week. Here's the exact 3-day system I use to run my company in 20 hours flat."
Why it works: Mirrors the source's pattern-interrupt ("Stop training every day") and applies it to founder burnout — the pain point your course solves.
📝 3. Adapted Script
[VISUAL CUE: Close-up, founder looking tired at laptop, harsh overhead light] Stop "hustling" 7 days a week. It's why you're stuck.
[VISUAL CUE: Hard cut — same founder, now energized, clean desk, soft natural light] Here's the exact 3-day system I built in Notion to run my company in 20 hours.
[VISUAL CUE: Screen recording — Notion dashboard with 3 colored blocks: Build / Sell / Rest] Day 1: Build. Day 2: Sell. Day 3: Rest. That's it. Everything else is noise.
[VISUAL CUE: Quick montage — calendar dragging, Slack closing, phone face-down] No context switching. No 14 open tabs. No "quick check" at 11pm.
[VISUAL CUE: Founder smiling, holding coffee, end card with course mockup] I packaged the whole system into a 7-day course. Link in bio — it's $49 and it'll buy you back 10 hours this week.
📱 4. Instagram Caption
The 3-day founder week that replaced my 7-day grind 👇
I used to "work" every single day and ship nothing. Then I built this Notion system. Now I run the whole company in 20 hours and actually take weekends.
7-day course. $49. Link in bio.
#solopreneur #notiontemplate #founderlife #productivityhacks #buildinpublic #indiehacker #notionsetup #startuplife #timemanagement #worklessdomore
🎥 5. Visual Direction
| Shot | Camera | Lighting | Notes |
|---|---|---|---|
| Opening | Tight close-up, eye-level, 35mm | Cold, overhead, slightly harsh | Sells the "stuck" feeling |
| Reveal | Medium shot, same angle, jump cut | Warm natural window light | Mirror the source's hard cut energy |
| System walkthrough | Screen recording, 1080x1920, 1.2x speed | N/A | Highlight blocks with subtle zoom-ins |
| Montage | Handheld, fast cuts (~0.4s each) | Matches reveal | Builds momentum |
| Close | Medium-wide, slight smile, end card overlay | Warm | CTA must appear by 0:32 to retain |
Aspect ratio: 9:16 · Length target: 35–40s · Captions: burned-in, bottom-third, max 5 words per frame.
Clean dashboard input |
Skeleton loading state |
Generated results (light mode) |
Generated results (dark mode) |
Note
Screenshots above are clean SVG mockups designed to match the actual UI exactly (rendered natively by GitHub). Swap them with real PNG/JPG captures any time — just drop them into /public/screenshots/ and update the extensions.
Hit a snag? Expand the relevant section below.
❌ yt-dlp: command not found
Recast shells out to yt-dlp to fetch videos. Install it for your OS:
# macOS
$ brew install yt-dlp
# Linux (via pip)
$ pip install yt-dlp
# Windows
> winget install yt-dlp.yt-dlpAlready installed but still failing? Set YTDLP_PATH in your .env to the absolute binary path (e.g. /usr/local/bin/yt-dlp).
❌ ffmpeg: command not found
ffmpeg is required for audio/video processing. Install it for your OS:
# macOS
$ brew install ffmpeg
# Linux (Ubuntu/Debian)
$ sudo apt-get install -y ffmpeg
# Windows
> winget install Gyan.FFmpegIf installed in a non-standard location, set FFMPEG_PATH in your .env to the absolute binary path.
⚠️ Gemini API quota exceeded
The Gemini free tier has rate limits (currently 15 requests/min and 1,500 requests/day for Gemini 2.5 Flash). If you hit them:
- ⏳ Wait a minute and retry
- 💳 Upgrade to a paid plan at ai.google.dev/pricing
- 🔁 Verify your key is correct and active in Google AI Studio
⚠️ Video download failed
This usually means one of the following:
- 🔒 The video is private or requires login
- 🌍 The video is region-locked in your area
- 🗑️ The video has been deleted or made unavailable
- 📡
yt-dlpis out of date — update it:$ pip install -U yt-dlp # or $ brew upgrade yt-dlp
Try a different public video URL to confirm your setup is working.
⚠️ Port 3000 already in use
Another process is using port 3000. Either stop that process, or run Recast on a different port:
$ npm run dev -- -p 3001Then visit http://localhost:3001.
Now
- Multi-platform support (Instagram, TikTok, YouTube Shorts, Reels)
- Dark mode with system preference detection
- Original language preservation in generated scripts
- Bulk URL processing (queue up to 10 videos at once)
- Export analysis & scripts to PDF / Notion
- Save analysis history with searchable dashboard
- One-click "Regenerate hook" button
Next
- A/B variant generation — 3 hook variants per script
- Brand voice profiles (upload past ads → match tone)
- Slack & Discord integration for team workflows
- Thumbnail concept generator (text-to-image via Imagen)
- CSV export for media-buying teams
- Custom prompt templates per niche (D2C, EdTech, SaaS)
Later
- Direct publish to Meta Ads Manager & TikTok Ads
- Multi-language UI (Hindi, Spanish, Portuguese, Arabic)
- Chrome extension — analyze any video without leaving the tab
- iOS & Android companion apps
- Team workspaces with role-based access
- Fine-tuned model for vertical-specific ad copy
Have an idea? Open an issue — the roadmap is shaped by the community.
PRs, issues, and ideas are all warmly welcome. Whether you're fixing a typo or shipping a new platform integration — you belong here.
1. Fork & clone the repo
git clone https://github.com/YOUR_USERNAME/recast.git && cd recast2. Create a feature branch
git checkout -b feat/your-amazing-feature3. Commit using conventional commits
git commit -m "feat: add bulk URL processing"4. Push & open a Pull Request
git push origin feat/your-amazing-featureNew here? Start with a [good first issue](https://github.com/thegauravmahto/recast/issues?q=is:open+is:issue+label:"good first issue") — they're handpicked to be friendly and well-scoped. Don't be shy, drop a comment and we'll guide you through.
Is my data private?
Yes. Videos are downloaded locally on your machine via
yt-dlp, sent to Gemini only for analysis, and deleted from local storage immediately after. Nothing is stored on a Recast server — there is no Recast server. Your API key, your data, your control.
How much does it cost to run?
For most personal use, nothing. Gemini 2.5 Flash has a generous free tier (~1,500 requests/day) that covers casual creators easily. Heavy users pay Google's pay-as-you-go rate — typically less than $0.01 per video analysis. No Recast subscription, ever.
Why did my script come back in Hindi / Tamil / Spanish?
That's intentional! Recast preserves the original language of the source video by default — a Hindi reel produces a Hindi script. This keeps cultural nuance and idioms intact. Want English output regardless? Toggle "Force English" in advanced settings.
Can I use this commercially?
Absolutely — Recast is MIT licensed. Use it for client work, agency workflows, internal tools, or paid SaaS. Attribution isn't required, but a link back or a ⭐ is always appreciated.
Does it support long-form YouTube videos?
Yes, up to ~1 hour. For videos longer than 5 minutes we recommend switching to audio-only mode — it's ~10× faster, ~5× cheaper, and quality drop is minimal for ad-copy generation since the hook intelligence lives in the script, not the visuals.
Why Gemini 2.5 Flash and not GPT-4o or Claude?
Three reasons: (1) Gemini has native video understanding via the Files API — no frame extraction hacks needed. (2) The 1M-token context window swallows long videos whole. (3) Cost-per-quality is unbeatable at this tier. That said, the architecture is provider-agnostic — swapping in GPT-4o or Claude is a ~50-line change in
lib/ai/provider.ts.
Recast stands on the shoulders of giants. Huge thanks to:
- Google DeepMind — for Gemini 2.5 Flash and the Files API
- Vercel — for Next.js 16 and effortless deployment
- shadcn — for the beautiful, ownable UI components
- yt-dlp team — the unsung heroes of video tooling
- FFmpeg team — for 20+ years of media wizardry
- Cal.com — for design inspiration and OSS-done-right vibes
- Lucide — for the crisp icon set
- Every contributor, issue-opener, and ⭐ giver — you make this fun.
Released under the MIT License — free for personal and commercial use. Build cool things.
