AutoDirector

One camera. Three angles.
Someone else does the switching.

AutoDirector listens to your podcast and cuts to whoever is talking — on a single 4K camera you already own. No config marathon. It just works.

Free while in public validation — no card, no signup

Watch it direct a real conversation

Unedited screen capture: setup wizard, voiceprint enrollment from one shared microphone, then the director running on a real two-person interview — the cuts you see are the cuts it made.

Download for Windows — Free

Windows 10/11 · ~124 MB · portable, no installer. Loading download count…

The app, for real

Director console while live — On air — the console follows the conversation and logs every cut.

Voiceprint enrollment — One shared microphone? Record each voice once — voiceprints tell people apart.

Bind voices to scenes — Click a person, click their shot. Bound in seconds — audio-only scenes are filtered out.

Director console — The console. Live thumbnails, cut history, one-key manual override.

Timing settings — Four timing knobs if you want them. Defaults are honest.

Connect to OBS — Connects to your OBS in one click — the plugin ships with the app.

How it works

Connect OBS

The app ships with an OBS bridge plugin. One click finds your OBS — your scene list and thumbnails appear automatically.

Point it at your voices

One mic per person, or a single shared mic — in which case you record each voice once and voiceprints tell speakers apart from then on.

Press Start

It listens and cuts to the current speaker, holds a wide shot when it can't tell, and steps aside the moment you touch a hotkey.

Measured, not marketed

Identical material, identical scoring, three-person dialogue testbench. Attribution accuracy / turn coverage:

Condition	Market leader ($69)	AutoDirector
Clean audio	100% correct	100% correct
Crosstalk −12 dB	80%	100%
Crosstalk −3 dB	27% of turns	86–100% of turns
One shared mic	0% — cuts wrong every time	stays on the wide shot, voiceprints decide

Test method and raw data are published in the open. Numbers you can't reproduce are marketing, not engineering. Read the shared-microphone measurement →

Requirements

Windows 10/11
OBS Studio 30+ (websocket enabled — the wizard turns it on)
One 4K camera — close-ups are crops of the same frame
Microphones: one per person, or one shared mic for voiceprint mode

Feedback — tell us what's missing

This build is free because we're validating it in the open. Whatever frustrates you, whatever you need before this replaces a human director — it goes straight to the founder's inbox.

Questions

Is it really free?

Yes — during the public validation phase the whole app is free, no expiry and no signup. When it graduates to a paid product there will be a one-time license; early feedback that ships a feature earns a free license.

Does it need the internet?

No. Everything runs on your machine and talks only to your own OBS on localhost. The single exception: one anonymous counter ping at launch so we can see how many people are using the build. It carries no content and no identifiers beyond a random install id.

What if it cuts wrong?

F1–F5 switch seats and wide manually. The AI stands down for 5 seconds and follows your lead. Works from a Stream Deck too.

Will my close-ups look soft?

The wizard checks your crop factors and warns you before they do. 4K → 1080p gives 2× zoom headroom; we refuse to silently upscale past it.

Can I use one camera for a two-person podcast?

Yes — that's the whole point. Shoot in 4K, AutoDirector crops left and right halves as two close-up angles plus the full wide. One tripod, one cable, zero extra cameras.

How does it know who is talking on a shared microphone?

Voiceprints. Each person records a 12-second sample once; after that the app matches voices in real time. On a three-person shared-mic testbench the market leader switched wrong 100% of the time — AutoDirector holds the wide and only cuts when a voiceprint says so.

Does it work with wireless mics or an audio interface?

Yes. Any source OBS can see — USB mics, wireless lavs, interfaces, or audio played into OBS — can drive the switching.

What can't it do?

It's built for seated conversations (podcasts, interviews, panels). People walking a stage need camera operators. Content-level calls (laugh → wide shot) are on the roadmap, not in this build.