Drop a video, open the camera, or pick a generative shader. STRUCTURE reshapes geometry, COLOR tones the image, GRADE fine-tunes hue and saturation, FX RACK adds feedback trails — and TRACK finds motion in anything.
Start simple: load a source, pick one STRUCTURE, choose a COLOR, then turn knobs until the canvas answers.
Shortcuts, signal flow, and where to start.
Preset digits are ignored while a knob has focus, so a stray keypress mid-tweak can’t snap your look.
Snap and Export need you to be signed in. Everything else runs entirely on your machine.
A dropped video, your webcam, a still image, or a generative GLSL shader from the library. Everything downstream treats them identically.
One full-frame pass that rebuilds the image as form — ASCII, dither, edges, halftone, mosaic. Its Output mode picks how that form is drawn: Source, Mono, Invert, or Color Isolate.
One selected effect across three tabs. Maps are pure per-pixel ramps. Unique effects build something new — atmosphere, light, motion response. Custom is ChromaEngine, where you author the ramp yourself.
Hue rotation and saturation, always available — even with Color set to none. Both knobs take modulation.
Up to three chained slots after the color stage. Feedback effects (trails, tunnels, burn-in) each keep their own buffer, so two slots running the same effect trail independently.
Detection runs beside the GPU pipeline, never through it. Blobs drive overlays, connection lines, track FX, and an optional per-blob synth composited on top of the final image.
Right-click any knob to attach a tempo-synced LFO. Modulation is a pure function of the clock, so what you see live is exactly what renders on export.
Drag a video onto the canvas, or open the Source section and pick a shader or your webcam. Nothing else lights up until there is something to look at.
In Synth mode, choose a Structure effect and set its Output mode. This is the biggest single change to how a frame reads.
Pick one Color effect, then push Hue and Sat in Grade until it sits right. Each effect remembers its own knob positions when you switch away and back.
Add up to three FX slots. Start with one — feedback effects compound fast.
Store the look in the preset grid, then recall it with 1–9. With a timeline segment selected, recalling a preset commits it straight to that segment.
Live mode gives you big targets, the preset grid, and a pop-out output window for a projector. Export renders offline at full quality.
The full pipeline — structure, color, grade, FX.
Detection, overlays, track FX, per-blob synth.
Performance surface: presets, shaders, output window.
Modes only change which controls are visible. Whatever you set stays running in the render.
The generative source library adapts published Shadertoy work — Star Nest by Pablo Roman Andrioli (MIT), Phantom Star by aiekick, the notebook-sketch pass by sofiene71 (CC BY-NC-SA 3.0), and others credited in source.
Object tracking runs MediaPipe Tasks Vision with EfficientDet-Lite, self-hosted so it works offline.