raea

Cut by what was said.

void reads footage the way an editor does, by its words, and writes the edit as code it can render, check and change.

2 percent of creative ▾
❯
CJ reelvideo
0:000:22

Both are the approved renders. The conversations are shortened; the code in Files is as void wrote it.

  1. 01

    Every word has a time.

    void transcribes each take word by word. CJ's 75-second take became 247 words, each with its start time, and cuts, captions and cards are placed on those words.

    media/golden/10-cj-reel/media/cj.words.json

  2. 02

    The edit is a file you can read.

    Each piece is a React composition written with @raea/motion. The reel's stat card is set to land when she says “two percent”, so it follows her if the cut moves.

    edit.tsx line 399: bigPop={cued("two percent")}

  3. 03

    It hears the mix.

    Music and sound effects are made for each piece and mixed under the voice. void measures the result: the CJ reel came out at −14.1 LUFS against a −14 target.

    measured with ffmpeg ebur128 on reference/approved.mp4

  4. 04

    Approved means a person watched it.

    Each reference piece was approved by a person, with sound, before it became the bar. New renders are compared with it frame by frame and on loudness.

    case.json: ssim_min 0.95, lufs_tol 0.5 · reference/APPROVED

What happens after you ask.

The CJ reel, from her raw take to the approved cut, the way raea made it.

  1. 1It reads the take.

    raea transcribes the footage word by word, with each word's start time. CJ's 75-second take became 247 timed words to cut and cue against.

    cj.mov · raw take · 0:31.9

    CJ's raw take, before the edit

    cj.words.json · 247 words

    Back29.39to29.55Apps29.63Flyer30.03They30.59found30.83that31.15just31.39two31.63percent31.87of32.19creative32.35ends32.99up33.23capturing33.39about33.87half34.11of34.35the34.51spend34.67and35.15the35.31top35.4710%35.71

  2. 2It writes the edit as code.

    The cut, the stat card and the captions are one React file. The card's “2%” is pinned to the moment she says “two percent”, so if the cut moves, the card moves with it.

    spaces/cj-reel/edit.tsx · line 399

    <StatCard from={89} y={236} h={440}
      big="2%" bigW={321} bigPop={cued("two percent")}
      lineSize={70} lines={[
        { text: "of creative", color: WHITE, at: 91 },
        { text: "captures", color: GREY, at: 93 },
        { text: "half the spend", color: LIME, at: cued("half") },
      ]}
      bars={[
        { label: "share of creative", value: "2%", at: 89 + 8 },
        { label: "share of spend", value: "~50%", at: cued("half") },
      ]} />

    cued("two percent") → 31.63 s in the take · cued("half") → 34.11 s

  3. 3It renders every frame.

    672 frames at 30 a second, on the Mac's GPU, and the same code renders on the iPhone. Change a line and only the frames that line touches render again.

    672 frames · 30 fps · 1080 × 1920

    Frame 1 of the reel
    Frame 2 of the reel
    Frame 3 of the reel
    Frame 4 of the reel
    Frame 5 of the reel
    Frame 6 of the reel
    Frame 7 of the reel
    Frame 8 of the reel
    Frame 9 of the reel
    Frame 10 of the reel
    Frame 11 of the reel
    Frame 12 of the reel

    edit: the card's bar · 2 of 12 shown frames render again

  4. 4It checks its own work.

    Before you see anything, raea looks at the frames, checks every caption against her voice, and measures the mix. This one came out at −14.1 LUFS, on target.

    the mix · 22.4 s

    −14.1 LUFS

    • ✓lookevery cut and card, frame by frame
    • ✓checkeach caption within one frame of her voice
    • ✓measure−14.1 LUFS integrated, target −14

Questions

Is void an AI video editor for Mac?
Yes. void is a Mac app: you describe the edit, and it transcribes your footage, writes the edit as code, renders it on your Mac's GPU and checks it before you see it.
How does text-based editing work in void?
Every word of every take gets a start time. Cuts, captions and cards are pinned to words, so when the cut moves, everything tied to a word moves with it.
Can it add captions automatically?
Yes. Captions come from the word-by-word transcript and are checked against the voice: in the approved reels, each caption lands within a frame of the word.
Does my footage leave my computer?
Rendering happens on your Mac's GPU. The footage is read and rendered locally.

The other ways in

What would you make first?

Tell void, and leave your email. Early access opens in small batches, and we'll write when yours is ready.

Early access opens in small batches.