bettervid
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote
bettervid turns a sentence, or your own files, into a finished, narrated video, for any purpose: a lesson from your notes, a paper turned into a feed of short explainers, a pitch, a history, an onboarding, a how-to. The problem: making a good explainer video takes a team (research, a writer, a director, a designer, an animator, an editor) and days of work, so most knowledge never becomes video. Template tools give everyone the same slides with a voice over. bettervid works like the whole team instead. How you use it: type what you want, or open Project mode, drop in PDFs, slides, spreadsheets or pictures on the left, ask for videos in the middle one message at a time, and scroll the finished videos on the right like a feed. Each video page has the video, Download and Copy link, with the script and its sources folded away. How it works: Muse Spark reads why you are asking, researches with web search when the topic needs it, writes the script in one pass, and directs the video moment by moment: where the eye goes, what to hold back, where to pause, how a demonstration plays. Then it writes every scene as text in our own scene language, which the engine compiles and renders with Remotion. Instruments measure what was drawn (how much of the frame the subject fills, legibility, where the eye goes, timing against the words). A reviewer and an art director look at the real frames and name exactly what to change, and Spark corrects its own scene. When a video demonstrates something exact, like a sort or a shortest path, a kit computes it, so every number on screen is right by construction. Muse Image draws the things a scene depicts, SAM cuts them into parts that move. Sound effects and music are cued to the scene's own moments and mastered to streaming loudness. What works today: the live site at https://bettervid.fly.dev makes videos end to end, with seven styles (blueprint, chalkboard, collage, flat, ink, neon, whiteboard) and Project mode. The gallery has the showcase videos, including two made from a real 17-page paper on education in Ethiopia. All code was written during the hackathon.