Kinematic 0.5.0 is out. Modeling and sculpting in the viewport, a timeline on every graph, photo and sound editing. What's new
Kinematic / Who it's for / Dance and performance

Let the dancer drive picture, sound and light.

A camera reads the body with no suit and no markers. Wrists, hips and hands become live values you can wire to anything in the piece: a projection, a moving light, a synth, a 3D character.

  • Choreographers
  • Dancers
  • Performance artists
  • Media designers
An ordinary webcamRuns on your own machineProjection, DMX and soundFree
points of light on the joints, tracked from a video of a dancer
the graph behind it
LocalAIImageToBodyPose→ImageTrail→ImageBloom→ImageBlend
The work

Make a dance piece in Kinematic.

From the first day in the studio to the tech, each part of the piece lives in the same file, and that file opens on any laptop in the rehearsal room.

  1. 01Rehearsal
    a skeleton over a dancer, read from a video file

    Track the dancer's body

    Point a webcam at the floor and the body pose node finds 17 points on each person, live. With tracking on, every dancer keeps the same number when they cross, so in a duet or a full company each body can drive its own part. The full body model reads 133 points, with the hands, the feet and the face.

    VideoCaptureInterface+LocalAIImageToBodyPose+LocalAIImageToWholeBodyPose
    Track a body →
  2. 02Visuals
    the left wrist sets the mirrors
    each frame fades into the next

    Turn movement into the picture

    Drag a tracked wrist onto any number in the app and it follows the dancer from then on. Here the left wrist sets how many mirrors the kaleidoscope has. Feed each frame back into the next and every move leaves a trail.

    LocalAIImageToBodyPose+ImageKaleidoscope+ImageFeedback+ImageBloom
    Let a dancer play the visuals →
  3. 03Sound
    Two tracked hands framing a rectangle of light that sets the synth
    two hands playing a synth

    Play the score with your hands

    Track both hands from the same camera. The gap between thumb and finger sets the pitch on one hand and the tone on the other, and the space between the hands sets the volume and reverb. A fist, an open palm or a thumbs up can fire a cue.

    LocalAIImageToHandKeypoints+LocalAIImageToHandGestures+AudioOscillator+AudioLowPass
    Play a synth with your hands →
  4. 04Projection
    a moving pattern lands on the dancer and nowhere else

    Project onto the dancer and nothing else

    A depth camera tells the dancer from the room. Capture the empty stage once, and anything that steps in front of it takes the picture, so the projector lights the body and not the back wall. Map the floor and the set with corner pins, grid warps and edge blends.

    DepthCameraCaptureInterface+Latch+ImageMath+ProjectionMapper
    Light only the dancer →
  5. 05Lighting
    a dancer on a 64 by 40 pixel grid, 40 universes

    Light the stage from the dancer

    Every pixel of a picture can be a light. Point it at the dancer and the LED fixtures carry the movement, one universe per Image to DMX node, sent out over Art-Net or sACN. A follow spot can find the dancer too, and steer a moving head after them.

    ImageToDmx+ArtNetSendInterface+DmxFollowSpotEngine
    Play video across an LED wall →
What Kinematic adds

Put the dance on a character.

Kinematic is a full 3D app, so the moves can leave the body. Read them live, from a suit or from a recording, and retarget them onto any rigged character. Here a salsa couple, recorded in a motion capture lab, is solved from its markers into two skeletons, and a moving head follows each dancer in the 3D preview.

3DGeoRigMarkersFromFile→3DGeoRigMarkerSolve→3DGeoRigClipEvaluate→3DGeoRigJointsToChannelData→DmxFollowSpotEngine
Put your dance on a character →
a recorded salsa couple, a follow spot on each dancer

Motion: CMU Graphics Lab Motion Capture Database (mocap.cs.cmu.edu), subject 60.

Record every run-through

Each take is named, numbered and saved as it ends, with the seconds before you pressed record. Save takes as BVH, FBX or GLB.

Live from one camera

The 3D body pose node reads 33 points in meters from a single webcam, and the body mesh node builds the whole figure.

From a suit or a lab

Take a performer from a mocap suit or a VMC app, or open a C3D file of markers and solve it into a skeleton.

On the night

Run the piece in Perform Mode.

Build the piece in the editor, with the picture live in the viewport and the graph beneath it. Press F1 and the editor steps aside: the piece fills the screen and goes to the projector you pick. Shift+Space pauses and resumes, and the picture holds through a pause.

Put the show on the big screen →
The editor playing a dancer video with points of light on his joints, and the eight nodes that make it on the canvas below
the editor: the piece in the viewport, its graph below
The same piece in Perform Mode: the dancer and the points of light fill the screen
F1: the same piece in Perform Mode

Open on the projector

Start straight into Perform Mode on the display you name. The editor stays out of sight.

terminal
$ kinematic piece.kin --perform --display 1

Controls on the other screen

Open the operator's control surfaces on the laptop screen while the piece plays on the projector.

terminal
$ kinematic piece.kin --perform --display 1 --controls-display 0

Keep it off the network

The tracking models download once. At the venue, seal the show off the network so nothing reaches out mid-piece.

terminal
$ kinematic piece.kin --perform --offline

Back on its feet after a crash

If the app ever falls over, it starts itself again, up to the number of times you set.

terminal
$ kinematic piece.kin --perform --restart-on-panic 3
How it fits

Gear, protocols and files.

Keep the cameras, the lighting desk and the sound system the venue already has. Kinematic sits beside them.

Cameras

  • Webcams and capture cardsIn
    Body, hands and face from an ordinary camera. Capture cards that show up as a webcam work the same way.
  • Depth camerasIn
    Orbbec Astra, Gemini and Femto, with color, depth and IR. RealSense-style cameras on Linux.
  • NDIIn and out
    Bring in a camera from across the room, or send your picture to another machine.
  • SyphonIn and out, macOS
    Share frames with other apps on the same Mac.
  • Video filesIn and out
    MP4, MOV, MKV and WebM play in. Record a run as H.264, H.265 or 10-bit ProRes.

Motion capture

  • VMCIn and out
    Take a performer from SlimeVR, VSeeFace, Warudo or VNyan, or send your character out.
  • Suits and optical systemsIn
    Adapter scripts that come with the app, for OptiTrack, Rokoko, Xsens, Qualisys, Axis Studio and Vicon.
  • C3DIn
    Labeled markers from a motion capture lab, solved into a skeleton.
  • BVH and FBXIn and out
    Takes and rigged characters, to and from other 3D and animation apps.
  • PosiStageNetIn
    Performer positions from a tracking system already in the venue.

Light and projection

  • DMX over USBIn and out
    Enttec DMX USB Pro, DMXking ultraDMX and other adapters, one universe per adapter.
  • Art-Net and sACNIn and out
    Many universes over the network.
  • Fixture profilesIn
    More than 1,500 fixtures in the Profile Library, drawn from GDTF and the Open Fixture Library.
  • Projectors and screensOut
    A window per output on any display, with corner pins, grid warps and edge blends.

Sound and show control

  • Microphones and sound cardsIn and out
    Record from a microphone or a sound card input, and play out to your speakers.
  • MIDIIn and out
    Keyboards, pads and control surfaces. Move a knob to map it.
  • OSC and QLabIn and out
    Let QLab cues fire scenes, looks and lights over OSC on TCP.
  • TimecodeLTC in, MTC in and out
    Lock the piece to the sound desk.
  • Phones and tabletsIn
    A touch page served from the show itself, or TouchOSC.
Where to start

Tutorials and demos.

Each tutorial starts from an empty file and takes about ten minutes. The demos come with the app: open one in the Demos browser, press Play, then take it apart on the canvas.

Tutorials

Demos

Not yet

What it does not yet do.

Better to know now. Missing something your piece needs? Reach out.

  • One skeleton from several camerasEach body tracking node reads one camera. Merging several camera views into one skeleton is not built in yet.
  • Kinect and ZEDThese depth cameras are not yet supported. Orbbec cameras are.
  • LasersThere is no ILDA or other laser output.
  • LTC outKinematic reads LTC but does not write it. MIDI timecode and Art-Net timecode go both ways.
  • SpoutSharing video with other apps on Windows is on the roadmap. NDI works on every platform today.

Download Kinematic.

Free. No account and no license key. Rehearse on a laptop and perform from the same file, on macOS, Windows or Linux.