Synthograsizer
A creative instrument for generative AI

Play prompts like a
synthesizer.

Synthograsizer is a browser-based creative suite for structured AI prompt engineering — with hardware controller support, live generative sketches, and real-time video streaming to Daydream Scope. Built for artists, VJs, and researchers who think with their hands.

50+
Templates
10+
Studio Tools
46+
Agent Personas
Variations
Gemini Veo 3 p5.js MIDI OSC WebRTC Daydream Scope Agent Orchestration Trace Observability
What it does

Prompt engineering, reimagined as an instrument

Most AI tools treat prompts like text boxes. Synthograsizer treats them like performable parameters — every variable a knob, every template a patch.

Structured prompt control

A D-pad interface turns every prompt variable into a navigable parameter. Shape ideas by ear, not by retyping.

  • Variable-driven template engine
  • Weighted value cycling per parameter
  • Live token preview and editing
  • Keyboard, mouse, or MIDI navigation

Hardware-first performance

Wire physical controllers directly into the creative loop. Every knob, fader, and pad becomes a mapping surface.

  • Web MIDI API — CC + Note bindings
  • OSC UDP bridge for cross-tool sync
  • Performance layout for live use
  • Hardware-synth aesthetic theme

End-to-end generation

From prompt to pixels to projector — text, image, and video models all live one click away in the same canvas.

  • Gemini text & image generation
  • Veo video studio with batch queue
  • Embedded p5.js generative sketches
  • Fullscreen Display window for OBS
Inside the App

Everything in one instrument

Synthograsizer is a single, deeply integrated creative environment. Every tool below is available inside the app — no switching tabs, no separate logins.

Synthograsizer

Launch

The instrument. D-pad prompt navigator with MIDI + OSC hardware control, 50+ templates, batch generation, and a live p5.js canvas — all in one browser tab.

Open app →

Agent Studio

Built-in

Multi-agent Gemini orchestration with 46+ persona presets, configurable turn-order routing, autonomous turn-taking, structured artifact emission, live token streaming, and a per-turn agentic trace viewer for debugging consensus and tool use.

Image & Video Studio

Built-in

Gemini image generation, Veo 3 video with batch queuing, Smart Transform, reverse-prompt analysis, and metadata extraction.

Glitcher Studio

Built-in

Full glitch-art engine embedded directly in the app — destructive pixel effects, non-destructive filter chains, selection tools, and live P5 canvas feed.

P5.js Canvas

Built-in

In-page generative sketch runner with live variable sync, CSS FX drawer, and External Display output for OBS and Daydream Scope.

MIDI & OSC

Built-in

Map any MIDI controller to any template variable. Route OSC to TouchDesigner, VDMX, or Max — every parameter is a performable knob.

Music Studio

Built-in

Lyria RealTime audio generation driven by the current prompt context — ambient scores, generative soundscapes, live to output.

Workflow

From idea to live output in four moves

Designed for the way creative work actually happens — iterative, tactile, and never locked into a single pipeline.

01 / COMPOSE

Load a template

Pick from 50+ included templates or import your own JSON. Generative sketches, cinematic prompts, character generators — all share the same variable format.

02 / CONTROL

Map your controller

Connect MIDI hardware or map OSC destinations. Every variable becomes a knob. Save the mapping per template for instant recall.

03 / GENERATE

Render in any medium

Send prompts to Gemini for text, image, or Veo video generation. Run a live p5.js sketch alongside. Batch 100 variations, or generate a single curated output.

04 / BROADCAST

Stream to the world

Push the output to OBS via the Display Window, stream the canvas to Daydream Scope over WebRTC, or route Spout to any visual pipeline.

Under the Hood

How the agents work

Agent Studio is a small orchestration runtime — the pieces that make a multi-agent conversation reproducible, inspectable, and steerable.

Turn-order routing

A configurable policy decides who speaks next — round-robin, mention-driven, or autonomous turn-taking with a stop condition. Per-template overrides keep behaviour predictable.

Per-turn prompt assembly

Each agent's prompt is composed at runtime from a persona, the shared transcript, and any artifacts emitted by prior turns. The same JSON template format defines personas and prompts.

Trace capture & consensus

Every turn writes a trace record — inputs, raw model output, parsed artifacts, latency. The trace viewer renders the run for debugging; consensus detection ends the loop when agents converge.

Template Format

One JSON file.
Infinite variations.

Templates are portable, readable, and designed to be authored by humans or generated by LLMs. A template is a prompt skeleton plus a set of weighted variable lists — optionally with an embedded p5.js sketch for live visual output.

The same JSON drives prompt assembly, batch evaluation runs of up to 100 variations, and agent persona definitions — one schema, three runtimes.

Import from PNG metadata, generate with the Template Generator, or author by hand. Share them, remix them, version them in git.

strange-attractors.json
{
  "promptTemplate": "a {{attractor}} in {{palette}}, {{mood}}",
  "variables": [
    {
      "name": "attractor",
      "feature_name": "Attractor Type",
      "values": [
        { "text": "lorenz butterfly", "weight": 2 },
        { "text": "clifford spiral", "weight": 1 },
        { "text": "rössler ribbon", "weight": 1 }
      ]
    }
  ],
  "p5Code": "p.setup = () => { p.createCanvas(800, 800); }; ..."
}
Integrations

Plugs into the pipeline you already use

Built to sit comfortably alongside existing creative software — not replace it.

Models & APIs
Google Gemini
Text + template generation
Veo 3
Video generation
ComfyUI (Legacy)
Local Stable Diffusion bridge
Creative Surfaces
Daydream Scope
OSC + WebRTC + Spout
Web MIDI
Any MIDI controller
OBS Studio
Browser Source + Spout output
p5.js Instance Mode
Live generative sketches
OSC UDP
TouchDesigner, VDMX, Max

Ready to play with prompts?

Open the suite in your browser. No install. No account. Just you, a template, and a canvas.