00:00 · in the feed by tonight

Raw clips in. A finished reel out.

Drop your takes — or just photos — and oriiion's editor cuts them, burns the captions, writes the hook, composes the music and levels everything to −14 LUFS. You publish; it edits.

Open the editor Get an access key

cuts on real pauses captions only when someone speaks music with no rights holder

c0-t0 · 6.2s"So here's the pricing mistake…" |cut| c0-t1false start ≈dissolve + whoosh≈ c1-t0 · 4.8s"Price was never the problem." |cut| photo-c2 · 4.0sslow push-in, ♪ bed
▸ takes cut on proven pauses — never mid-word ♪ music bed −14 LUFS, composed by the worker ≈ transitions land in silence, not across speech

00:04 · what the editor does

It edits the way an editor does. It just doesn't sleep.

Every job is measured before it is cut: what was said, where the pauses are, what moves on screen, whether anyone is speaking at all. The decisions below happen on every render — none of them need you to know what an EDL is.

edit is 7/11 take(s), 31.5s

Auto-cut from your takes

Transcripts are split into takes at real pauses. The editor keeps the strongest, drops false starts, and can lead with a line you said last.

captions still, below the head

Captions that dodge the face

Vision finds the speaker's head and places subtitles around it, inside every platform's safe area. Ten styles, your brand color, any Google font.

3 still image(s) → 4s clip(s)

Photos become footage

Drop images alongside clips — or alone. Each becomes a few seconds of slow push-in, at its own aspect ratio, ready to cut like anything else.

soundtrack mixed — generated music bed

Music it composes itself

Silent footage gets an ambient bed the worker writes with pure synthesis — no packs, no licensing, and the same music on every re-render.

hook placed 0.29, cover cached

Hook, CTA and cover

A title written from what you actually said (or your context), an end-card that fits the video's goal, and a cover frame picked by vision.

cleaned and levelled to −14 LUFS

Broadcast loudness, always

Every job is denoised and levelled to the loudness Instagram, TikTok and YouTube normalise to. Not a setting — a property of the pipeline.

00:12 · burned in, not bolted on

Ten caption styles. One of them is yours.

This is default

default · Anton

MAKE IT POP

hormozi · Archivo Black

ONE

single-word-box

READ ALONG NOW

karaoke highlight

00:19 · it listens before it writes

Captions only when someone actually speaks.

Transcribers answer even when nobody spoke — half-heard lyrics over a music bed, "thank you for watching" over silence. The worker judges every clip's audio first, so what gets burned in was actually said.

speech

Someone is talking. Transcript, captions, hook and CTA are written from what they said.

sound-only

Music, traffic, a dog. No invented captions — and no music bed added on top of yours.

silent

No audible audio. No captions, no transcription cost — and a composed bed so it never ships as dead air.

00:27 · access

oriiion users are already in.

If your team uses oriiion, your access comes with it — the app provisions your key and it stays valid for as long as you're an oriiion user. Nothing to set up.

Not on oriiion? Ask your admin for an invite: they create your account with one request and hand you a personal access key. Paste it once in the editor and you're in. Lost it? Keys rotate in seconds; they're never stored readable.

Open the editor 
# Invite a user (admin, one request)
curl -X POST https://video.oriiion.tech/users \
  -H "Authorization: Bearer $SERVICE_KEY" \
  -d '{ "email": "anna@studio.se" }'

# → { "apiKey": "wk_…" }  shown once, hashed at rest

# Render with it — same API the editor uses
curl -X POST https://video.oriiion.tech/jobs \
  -H "x-api-key: wk_…" \
  -d '{ "videoUrl": "…", "autoEdit": true }'