Private by design
Faces are analyzed in your browser, on your computer. Your video frames are never sent to a server to be analyzed.
Your cameras, directed by AI
The AI Director watches every camera, reads who is looking at the lens, the talking gesture and where heads turn, and cuts to the best shot at the pace of your show. Hands free, right in your browser.
How it decides
It watches every face, on every camera
Several times a second, each camera gets a score from what its face shows. When another camera clearly leads, and the current shot has run long enough for your show, it cuts.
Gaze
Iris and face landmarks tell who is looking at the lens, the strongest signal of who the show is about right now.
Talking gesture
The talking gesture is read from the face, not the microphone, so one shared studio mic never confuses which camera is speaking.
Head pose
Where heads turn, in 3D. A guest turning to answer the host is a cue to cut.
A real session · Two cameras
A real session, not a simulation: the host switches between his webcam and his phone, and the AI Director cuts on its own. The mesh and the scores are what MediaPipe Face Landmarker, the model the AI Director runs, measures frame by frame on each camera's full frame.
Reads every camera slot
Each connected camera is analyzed at its full resolution, several times a second.
Sees who looks at the lens
The face mesh, the iris and the head axes give each camera its score.
Notices the turn
When the host looks away, that camera's gaze and head scores drop.
Cuts to the right camera
The camera the host now looks at wins and goes to Program.
Measuring CAM 1
- Gaze to lens
- 0.69
- Head toward camera
- 0.89
Pacing
Cuts at the pace of your show
A calm podcast and a fast gaming stream should not cut the same way. Pick a show type and the AI Director tunes how long it holds a shot, how sure it must be and how much each signal counts.
4 s
Shortest time between cuts
Calm, stable switching for seated conversations.
The most it could cut in 20 seconds
Auto-TuneNot sure which one fits? Auto-Tune recognizes the kind of show and adapts the timing as you stream.
A director that knows its place
It runs the switching so you can run the show, and it never gets in your way.
Auto-Tune
Machine learning that recognizes podcasts, interviews, panels and action, and adapts timing and weights as you stream.
Your wide shots stay yours
Inputs with camera layers become presets that the AI skips, so a composed wide shot is only used when you choose it.
Take over anytime
Turn it on or off with one click or a MIDI button. While it runs, keyboard shortcuts pause so nothing fights for the switcher.
Webcams, phones and guests
It analyzes local webcams and WebRTC cameras: your phone connected by QR code and remote guests.
Scales with your cameras
Analysis runs in parallel background workers, from 1 to 8 depending on your plan, so more cameras stay responsive.
Questions
Does it listen to my audio?
No. It reads faces only: gaze, head pose and the talking gesture. One studio microphone can serve many cameras, so audio cannot tell which camera is speaking.
Is my video sent anywhere to be analyzed?
No. The analysis runs in your browser, on your computer. Only the detection model is downloaded.
How many cameras do I need?
At least two active inputs. Inputs with camera layers are kept as manual presets and are not counted.
Which plans include it?
Plans with two or more inputs, starting with Beginning. Higher plans run more analysis workers in parallel.
Can I take over during the show?
Yes. Turn it off with one click or a MIDI button and switch by hand. Turn it back on whenever you want.
What does it need from my computer?
A modern browser. On lighter hardware, lower the analysis resolution or frame rate in the Advanced panel.
Explore more of the Studio
Hand the switcher to the AI
Connect two cameras, turn on the AI Director and focus on your show.
Start free