FOUR TAKES — The AI Captioning Studio

Every AI captioning tool does video-in, caption-out. FOUR TAKES is the only one that ships captions PRE-JUDGED: an internal LLM judge scores every caption on accuracy and tone before you see it, and anything below 8/8 is automatically retaken with a concrete fix instruction. Brand-safe by construction. HOW IT WORKS 1. SAMPLE — ffmpeg pulls 16–32 frames evenly across the FULL clip (a user-facing Sampler slider trades speed vs accuracy). 2. TRANSCRIBE — Whisper v3 on Fireworks captures speech, in parallel with sampling. 3. DOSSIER — Gemma (multimodal) compiles a strict facts-only scene dossier: subjects, actions in order, on-screen text. Nothing invented. 4. FOUR TAKES — Gemma writes all four styles concurrently from that ONE dossier, so every voice agrees on the facts while the tone transforms. Each style is a calibrated persona (The Anchor, The Cynic, The Debugger, The Standup) with a voice spec, few-shot anchor, and explicit avoid-list. 5. JUDGE — scores each take 0–10 on accuracy and tone; failures get up to two retakes, shown live in the UI. Gemma · Fireworks AI · Whisper · ffmpeg · FastAPI · Docker. Crafted by The Maker (Rahul D).