Computer vision / 2026 / Runs natively
Video FX Lab
Python renderer for the blue + red → hollow purple sequence, running either on fixed timings or as a live hand-driven state machine.
01
The problem
Matching an effect to reference footage means committing to either fixed timing or live control.
02
How I approached it
Two modes over one renderer: a timeline mode following phase timings analysed from the reference clip, and an auto mode driven by live hand landmarks per frame.
03
The outcome
The same effect usable for rendering and for live performance.
Related work
A webcam filter that reads your face and hands, picks a reaction pose, and stamps the artwork over your head, then publishes it to a virtual camera Meet, Zoom, Teams and Discord can select.
ASL fingerspelling to text from a webcam. It reads hand geometry instead of pixels, so all 26 letters work with no training, and two seconds of calibration per letter fits it to your own hand.
Hand-tracked cursed-technique visualiser: MediaPipe landmarks drive a Three.js particle engine, with an ML classifier picking the technique in real time.