Creator Lab

Watch the full-quality demo · Download MP4 · Saved analysis replay

Turn a creator's Instagram Reels into a searchable research library. Filter by topic and hook, compare engagement, read the scripts, and open the original posts behind each pattern.

Built with Apify → Fireworks or Groq → TypeSafe Jev. Runs locally in your browser. Bring your own API keys and choose a public creator, including your own account.

Full setup guide covers installing the tools, getting keys, choosing a transcription provider, your first analysis, and troubleshooting.

1. Install Node.js (24 recommended; minimum 22.9) and FFmpeg. Both ffmpeg and ffprobe must be on PATH.

2. Download this repository using Code → Download ZIP, unzip it, and open a terminal in that folder. Or clone it:

   git clone https://github.com/artemnovitckii/creator-lab.git
   cd creator-lab

3. Create your local configuration:

   npm run setup

4. Open .env in your editor. Add APIFY_TOKEN, TYPESAFE_API_KEY, and one transcription key: FIREWORKS_API_KEY or GROQ_API_KEY. Set TRANSCRIPTION_PROVIDER to match.

5. Check and launch:

   npm run doctor
   npm start

6. Open http://127.0.0.1:5190, verify Connections, then choose New analysis. Enter a username without @. Start with a 20-Reel pilot.

No npm dependencies or build step are needed. You can open the synthetic motion rehearsal without keys; real collection and analysis use paid provider APIs.

• Eight transcript classifications: topic, opening move, hook mechanism, script structure, evidence, emotional appeal, advice specificity, and spoken CTA.
• Script passages labeled as hook, setup, problem, example, advice, payoff, CTA, or other, with source text and available timestamps.
• Combined topic and hook filters, engagement comparisons with sample sizes, and original-Reel links.
• A synchronized thumbnail wall and performance map, with saved-result replay for screen recording.
• Pausing, resuming, transcript reuse, classification caching, and JSON exports.

Jev sees the speech before performance metrics are joined. Views and plays remain separate. Unknown metrics stay unknown. Engagement comparisons describe the selected sample; they do not prove what caused a Reel to perform.

Use either transcription provider. There is no automatic switch that could charge a different provider. Restart the server after changing .env.

You pay the providers directly. The Apify spending cap in the form covers Apify only, not Jev or transcription. Start small and check provider billing before increasing the batch. Displayed model costs are estimates using rates in the source, not a billing guarantee.

Each run requests 1 to 1,000 Reels from one handle. Actual coverage depends on Instagram and Apify. The scraper skips pinned and trial Reels; this is not a guaranteed full-account archive. Private, deleted, expired, or inaccessible media may fail. Music-only and very short speech are excluded from script analysis.

Replay animates saved results. It does not run collection or inference again, and its playback speed is not the pipeline's processing speed.

The server binds to 127.0.0.1. Keys stay in your local .env or server memory and are used to authenticate requests to their providers. Apify receives the handle; your chosen speech provider receives audio; Jev receives transcript text. Original source media is fetched from supported CDN hosts. The dashboard also loads fonts from Google Fonts. Full data flow.

.env, generated data, caches, and downloaded media are excluded from Git. This repository includes no personal API keys or collected creator archive. Never put keys into a GitHub issue or share your .env.

npm run check
npm test

Tests use mocked provider responses and local FFmpeg fixtures; they do not make paid API calls. Provider key verification checks account access, not a successful end-to-end analysis. A small live pilot is the final setup check.

• server.mjs: local HTTP server, API, credentials in memory.
• lib/providers.mjs: provider calls and media extraction.
• lib/pipeline.mjs: processing, persistence, retries, and cache.
• lib/schema.mjs: Jev classification definitions.
• public/: research dashboard and recording views.
• data/: generated automatically, local only.

MIT licensed. This is an independent project, not an official Instagram, Apify, Fireworks, Groq, or TypeSafe product.