Reel Engine
How NexusPoint Built a Pipeline That Turns Static Infographics Into 40-Second Motion Reels With Synced Voiceover
Client
Challenge
Instagram Reels are the highest-reach format on the platform 55% of views come from non-followers, and the algorithm prioritizes watch time, sends, and saves over likes. But for a founder running an AI agency, creating a reel that looks professional and sounds human is a multi-tool nightmare. The default approach is to record a talking-head video on your phone, throw it into CapCut, add auto-captions, and post. The problem is that static infographic posts which NexusPoint has 130+ of die on Reels. An infographic that performed well as a static image gets zero traction as a video. The content is good, but the format is wrong. Four specific problems: Static infographics underperform on Reels. The format expects motion, text-on-screen that arrives in time with the voice, and visual variety across scenes. An infographic as a single static frame gets swiped past in under 2 seconds it doesn't use any of the 5 ranking signals the algorithm measures. Voiceover sync is manual and brittle. Recording a voiceover in ElevenLabs, importing it into a video editor, aligning captions word-by-word, and adjusting scene timing to match the audio is a 2-4 hour workflow per reel. One phrase change means re-doing the alignment. No QA gate between transcription and render. Whisper mishears words ("a i" -> "AI", "car work" -> "Cowork", "codcodes" -> "Claude Code's"). Without a ground-truth correction step, every rendered reel has wrong captions and off-sync scenes. Catching those after a 45-minute render is expensive. Music mixing is easy to get wrong. Adding background music without accounting for the voiceover's natural pauses means music swells in the gaps between sentences the classic "weird sound before the next word" problem. Most auto-mix tools don't duck during interior silence gaps. NexusPoint built the Reel Engine to solve all four a Remotion-based pipeline that turns a content.json script into a fully synced, branded, music-mixed reel with a QA gate between every stage.
Goal
The Reel Engine is a Remotion 4.0 project at projects/reel-engine/ that turns an infographic post (content.json + voiceover audio + infographic image) into a 9:16, 40-50 second motion-graphics reel with branded animation, synced captions, and background music. It is not a video editor. It is a pipeline with 5 sequential stages, each with its own validation step: Write content.json (script + scene definitions) -> Generate or record voiceover -> Transcribe and align (Whisper -> captions -> timeline) -> QA gate (ground-truth correction + audio surgery + sync check) -> Render (Remotion -> mp4) + optional music mix (ffmpeg with intelligent ducking) Every stage gates the next. No partial output ever renders.
Result
The Reel Engine is a 5-stage pipeline that turns any infographic post into a fully synced, branded, music-mixed motion-graphics reel. Every stage is gated by a validation step that prevents defects from reaching the final video. 22 reels produced across 15+ topics. The architecture a content.json script that defines both scenes and voice text, a ground-truth invariant that makes alignment deterministic, a QA gate that auto-fixes Whisper transcription errors before they corrupt the render, and a music mixer that ducks during voice-silent gaps is not specific to AI agency content. The same system works for: Any content brand turning static infographics, carousels, or data visualizations into motion reels for Instagram and LinkedIn A product team creating demo reels that show product screenshots with synced voiceover walkthroughs An educational creator producing short-form lesson videos where captions must match spoken text exactly A marketing agency producing branded social videos at scale without a video editor on staff Any business sitting on a library of static visual content that underperforms on Reels but has no pipeline to motion The key innovation is the QA gate between transcription and render. By rebuilding captions from the ground-truth script (instead of trusting Whisper's output) and verifying sync before rendering, the pipeline eliminates the most expensive error in reel production: a rendered video with wrong captions that needs a 45-minute re-render.