Desktop apps for generative art and creative coding

A rig doing the rounds this month appears to pour LED pixels out of the bottom of a monitor and into a pile on the desk, with the light spilling off the screen and continuing down a strip of addressable LEDs in perfect sync. It looks like a hardware trick. It is mostly a software one: the same patch that renders the falling particles on screen also samples that render and pushes the matching pixel values out to a controller. Projects like that always start in a creative-coding environment, and picking the wrong one costs weeks. We tested eight of the best apps for generative art on desktop, spanning code-first sketching, node-based patching, pure shader work, and real-time output to physical hardware, on Windows, macOS, and Linux.

What to look for in a generative art app

Quick comparison

App Best for Platforms Free option Paid from Hardware output
Processing Learning to code visuals Windows, macOS, Linux Yes, fully free Free Art-Net and serial via libraries
Cables Node patching without cost Windows, macOS, Linux, web Yes, MIT licensed Free MIDI and OSC, no native DMX
Blender Procedural 3D and print-res output Windows, macOS, Linux Yes, fully free Free None
TouchDesigner Installations and LED mapping Windows, macOS Non-commercial, capped at 1280x1280 Around $300 education, around $900 commercial DMX, Art-Net, sACN
KodeLife Live GLSL and Metal shader work Windows, macOS, Linux, Raspberry Pi Yes, free public beta Optional one-time licence Audio and MIDI input only
openFrameworks Heavy C++ realtime work Windows, macOS, Linux Yes, open source Free Art-Net and serial via addons
Notch Touring and broadcast graphics Windows Trial that cannot save Around £99/month plus VAT Via media server integration
vvvv gamma Windows show control pipelines Windows Free for non-commercial use Commercial licence required Art-Net and DMX, 256 universes

The 8 best apps for generative art on desktop

1. Processing – best starting point for code-driven sketches

Processing is the tool most generative artists learn first, and a lot of them never fully leave. It wraps Java in a sketchbook editor where a blank file plus a setup() and draw() pair gets you an animating window in about a minute. The library ecosystem is where Processing for generative art gets serious: geometry, physics, computer vision, OSC, and Art-Net over DMX are all a contributed library away, and the sketch you wrote to fill a canvas with noise fields can drive an LED strip through a serial connection to an Arduino or Teensy without changing much of the drawing code. Sketches export as standalone applications for Windows, macOS, and Linux, which matters when a piece has to run unattended in a gallery.

Where it falls short: it is Java under the hood, so garbage collection pauses show up once you push particle counts high. Shader work is possible through PShader but it is a second-class path compared with tools built around GLSL, and there is no timeline, no scene graph, and no built-in way to schedule a show.

Pricing:

Platforms: Windows, macOS, Linux

Download: processing.org · GitHub

Bottom line: start here if you want to understand what your work is actually doing. Move on when frame times stop cooperating.

2. Cables – best free node-based environment

Cables is a patching environment where you connect operators into a graph and watch WebGL output update as you drag. It ran browser-only for years, which put people off for serious work, but the standalone build changed that: it is an Electron application for Windows, macOS, and Linux that works offline, reads local files and assets without upload limits, and can export a patch as an executable that runs on any of the three platforms. The whole project is MIT licensed, so nothing about using Cables for generative art is gated behind a tier, and patches you build stay yours to inspect and modify.

Where it falls short: it renders through WebGL, so the performance ceiling sits below tools that talk to the GPU natively. Large patches get hard to read without careful use of subpatches, and there is no native DMX or Art-Net output, so hardware work needs a bridge over OSC or MIDI to something else.

Pricing:

Platforms: Windows, macOS, Linux, web

Download: cables.gl · Standalone · GitHub

Bottom line: the best free way to find out whether node patching suits you, and good enough to finish screen-based pieces in.

3. Blender – best for procedural 3D and print-resolution output

Geometry nodes turned Blender into a generative tool that happens to also be a 3D suite. You build geometry from a graph of nodes, scatter and instance across surfaces, drive everything from a seed value, and simulation zones let one frame feed the next, which is how you get growth, flocking, and accumulation without writing a solver. Blender for generative art has one advantage nothing else here matches: because it renders offline through Cycles as well as in real time through EEVEE, the same setup that previews at 60fps can output a 10,000 pixel wide frame for print or a clean animation for video.

Where it falls short: it is not a live tool. There is no straightforward path for audio-reactive input, no DMX output, and nothing resembling show control, so it is the wrong choice for anything that has to respond to a room. Geometry node graphs also get verbose fast, and reading someone else’s setup is genuinely hard work.

Pricing:

Platforms: Windows, macOS, Linux

Download: blender.org

Bottom line: the pick when the output is a finished image or film rather than a running installation.

4. TouchDesigner – best for installations that drive LEDs

TouchDesigner is what the LED-pouring-out-of-a-monitor genre of project usually runs on. It is a node environment split into operator families for textures, channels, geometry, and components, and the hardware side is native rather than bolted on: you arrange your LED positions as points in 3D space, use a mapper component to project video onto those points, and send the result out through the DMX Out CHOP over Art-Net or sACN to a controller like a PixLite. Python is embedded throughout, so anything the nodes cannot express you script. It is the reason TouchDesigner for generative art dominates the interactive installation and touring visuals world.

Where it falls short: the free Non-Commercial licence limits image resolution to 1280x1280, which is fine for learning and a real problem the moment you want a 4K projector feed. There is no Linux build, only Windows and macOS, and macOS lags on GPU features. The operator model also takes a while to click, and the first month is mostly spent learning which family of node does what.

Pricing:

Platforms: Windows, macOS

Download: derivative.ca · Docs

Bottom line: the default for physical installations. Budget for the commercial licence if anyone is paying you.

5. KodeLife – best pure shader sketchpad

KodeLife does one thing: it puts a GLSL or Metal shader on the left and its output on the right, and recompiles as you type. There is no project structure to fight, no node graph, no build step. It supports every OpenGL GLSL flavour the host machine offers plus the Metal Shading Language on Apple hardware, and it takes audio input, MIDI, and gamepad data as uniforms, which is enough to make a shader react to a live set. It runs on Windows, macOS, Linux, and Raspberry Pi, which makes a Pi plus a small screen a plausible way to leave a shader running in a room for months.

Where it falls short: it is a shader editor, not an environment. No geometry pipeline, no timeline, no compositing, no file output beyond captures. It also assumes you already write GLSL, and if you do not, the learning curve is the shading language itself rather than anything about the app. The desktop builds are still labelled a public beta.

Pricing:

Platforms: Windows, macOS, Linux, Raspberry Pi

Download: hexler.net

Bottom line: the fastest way to iterate on a shader. Pair it with something else for anything that needs structure.

6. openFrameworks – best for work that has to run fast

openFrameworks is a C++ toolkit rather than an application. You write your sketch in whatever editor you prefer, compile it, and get a binary with no runtime overhead between your code and the GPU. That matters when the piece involves millions of points, multiple camera feeds, or a machine that has to run for a year without a restart. The addon ecosystem covers most of what installation work needs, including Art-Net for sending DMX data straight from your app, and the current 0.12 line supports Windows, macOS, and Linux with the usual project generator to set up builds.

Where it falls short: the toolchain is the tax. Setting up compilers, resolving addon dependencies that were last touched years ago, and debugging a linker error are all part of the deal, and none of it is drawing. Development has slowed to steady maintenance rather than rapid releases, and some addons have quietly rotted.

Pricing:

Platforms: Windows, macOS, Linux

Download: openframeworks.cc · GitHub

Bottom line: worth the setup pain when performance is the constraint. Skip it if you are still exploring what the piece is.

7. Notch – best for professional real-time production

Notch sits at the top end of the market, where the output goes to an arena screen, a broadcast studio, or an LED volume. It is a node-based real-time tool built around 3D scenes, particle systems, deformers, and effects that render live rather than to a file, and its position in touring production comes from how it integrates with media servers so operators can adjust parameters during a show. If your generative art work is heading toward commercial production, Notch is the tool the people already doing that work use.

Where it falls short: it is Windows only, requires a subscription rather than a one-time purchase, and the trial version cannot save your work at all, which makes evaluating it awkward. Entry pricing sits around £99 per month plus VAT, with the Pro tier roughly double that, so it is hard to justify unless the projects are paying for themselves.

Pricing:

Platforms: Windows

Download: notch.one · Pricing

Bottom line: for people billing clients for real-time graphics. Everyone else should spend the money on a better GPU.

8. vvvv gamma – best overlooked option for hardware pipelines

vvvv gamma is the one most people skip, and the one that solves a specific problem better than anything else here. It is a visual live-programming environment for .NET where patches compile to real code, which means the graph you draw runs at native speed and can call any .NET library you point it at. Lighting output is a first-class feature rather than an addon: Art-Net nodes ship with the software, and it handles the full Art-Net II range of 256 universes across 16 subnets, which is a lot of pixels. For a Windows machine that has to drive both a screen and a large lighting rig from one patch, vvvv gamma for generative art is a genuinely strong answer.

Where it falls short: Windows only, with no macOS or Linux plans worth counting on. The community is smaller than TouchDesigner’s, so there are fewer tutorials when you get stuck, and the VL patching model is unusual enough that experience with other node tools does not transfer cleanly. It is free for non-commercial, educational, and evaluation use only, so paid work needs a licence.

Pricing:

Platforms: Windows

Download: visualprogramming.net · vvvv.org

Bottom line: the right call for Windows show-control and lighting work. Not worth the learning investment for screen-only pieces.

How to pick the right one

FAQ

What is the best free software for generative art? Processing for code-driven work, Cables for node-based patching, and Blender for procedural 3D. All three are fully free with no capability limits, unlike TouchDesigner’s Non-Commercial licence, which caps resolution at 1280x1280 and forbids paid work.

Do you need to know how to code to make generative art? No. Cables, TouchDesigner, vvvv gamma, Notch, and Blender geometry nodes are all node-based, so you build logic by connecting operators instead of writing statements. You will still need to understand loops, conditions, and coordinate systems, because the concepts are the same whichever way you express them.

Is TouchDesigner free? The Non-Commercial licence is free and covers personal projects and learning, but it limits image resolution to 1280x1280 and cannot be used for any work you are paid for. Education licences run around $300 and commercial licences around $900.

Which software can drive LED installations? TouchDesigner and vvvv gamma both output DMX over Art-Net and sACN natively, which is what LED pixel controllers expect. Processing and openFrameworks can do it through Art-Net libraries and addons, or by talking to a microcontroller over serial. Blender and KodeLife cannot.

How powerful does your GPU need to be? For Processing sketches and Cables patches, most integrated graphics from the last few years will keep up. Raymarched shaders in KodeLife, heavy particle systems in TouchDesigner, and anything in Notch want a dedicated card, and high resolutions multiply that requirement quickly. Blender’s Cycles renders will use whatever you give them.

Is generative art the same as AI image generation? No. Generative art means writing rules or building systems that produce the image, so you control the process and can rerun it with different parameters or seeds. AI image generators produce output from a trained model and a prompt, which is a different practice with different tools.