A camera-interactive installation Intermediate

Objective — build a webcam installation where a viewer's movement spawns glowing particle trails, ready to project onto a wall.

What you'll see — a dark, moody field of particles that light up and trail wherever motion happens in front of the camera. Stand still and it calms; move and the room paints itself.

The optical-flow particle field driven by a moving test clip — with your webcam, the particles churn wherever people move. Captured live from the recipe's own output.

Before you start

• tdmcp installed for your AI client.
• The TouchDesigner bridge step done, and bridge running in TouchDesigner's Textport.
• A webcam. No webcam handy? Every prompt below can use a synthetic test source instead, so you can build and rehearse the whole look offline.

No Kinect required

This uses a plain webcam and optical flow (motion detection). For depth cameras and full-body pose, see MediaPipe adapters and Physical installations.

Steps

Copy each prompt into your AI client, one at a time. Wait for each to finish before sending the next.

1. Confirm TouchDesigner is connected and see your webcam:

   Check TouchDesigner is connected, then show me my webcam source. If no camera is available, use a synthetic test source instead.

   → The AI confirms the bridge is live and brings up a camera (or test) image.

2. Build the motion-reactive base from the recipe:

   Apply the optical_flow_particles recipe, driven by my webcam. Use the bundled test clip if my camera is not ready.

   → You get a network that reads motion from the camera as an optical-flow field and feeds it straight into the particle render — motion in front of the camera already stirs the particles out of the box.

3. Make the reaction stronger and driftier:

   Make the particles react more to movement and drift once they're stirred, so motion in front of the camera really paints them.

   → The particles respond harder to motion and keep drifting after they're pushed, so the field feels alive rather than snapping back.

4. Set the installation mood:

   Add trails to the particles and give it a dark, moody palette — deep blues and violet on near-black.

   → Particles leave glowing trails and the scene reads as a gallery piece, not a technical demo.

5. Expose the two controls you'll actually perform:

   Expose a Flow-sensitivity control and a Trail-length control so I can tune how reactive and how smeary it is.

   → Two live knobs appear. Flow-sensitivity sets how easily motion triggers particles; Trail-length sets how long the trails linger.

6. Preview and take it fullscreen for the installation:

   Show me a preview. Then tell me how to send this output fullscreen to my projector or second display for the installation.

   → You see the result, plus steps to route the output to a projector.

Expected result

A left-to-right network: camera → optical-flow field → particle render → output, with a preview showing dark particles that ignite into glowing trails wherever someone moves. Flow-sensitivity and Trail-length are exposed as live knobs. Sending the output fullscreen turns any wall into the piece.

If it goes wrong

• Webcam not found / black image → ask for a synthetic test source, then swap the camera back in later. On macOS, grant camera access — see the camera permission note.
• Too noisy — particles fire constantly → lower Flow-sensitivity, or say "only react to bigger movements."
• Nothing reacts → raise Flow-sensitivity, and make sure there's enough light and actual movement in frame.
• Projector / fullscreen setup → see Physical installations for reliable projector, calibration and room-sensor workflows.
• Still stuck? → Troubleshooting and the FAQ.