Lingdong Huang is a PhD student in the Future Sketches group at MIT Media Lab. In August 2025, he released Dither — a work-in-progress creative coding language. Code written in Dither can be interpreted (for fast iteration) or compiled (for fast execution); it runs on CPU or GPU; it targets the browser or native apps — one codebase, all substrates. The language has type inference, built-in vector and matrix math, and a standard library whose core job is getting pixels onto your screen as fast as possible. In April 2026, Huang released v0.0.3: an 8MB Mac IDE bundling the VM, compiler, REPL, debugger, standard libraries, and documentation in a single binary.

To see what Dither does structurally, you have to first see the already-construct it acts against. Creative coding practice has hardened, over twenty-odd years, into two parallel but non-communicating traditions: the CPU tradition (Processing, p5.js, openFrameworks — Java/JavaScript, sequential execution, readable, slow on complex rendering) and the GPU tradition (GLSL, WGSL — C-like, mandatory type declarations, parallel execution model, fast, but demanding a fundamentally different way of thinking). This bifurcation is not an incidental engineering choice that anyone made on purpose. It solidified on its own — two separate communities, two separate learning curves, two epistemologies that can't talk to each other. A p5.js sketch cannot become a GLSL shader without a complete rewrite. A shader artist can't iterate interactively without learning a separate framework. The split became construct: inherited, taught, assumed, treated as the starting condition for every new tool built afterward rather than as something to question.

Dither's chisel-construct move is to demote this split to a backend detail. The same Dither source code compiles to CPU or to shader; runs in the browser or as a native binary. Type inference means you don't write vec3 color = vec3(1.0, 0.5, 0.2) — you describe what you want, and the language infers the types. Vector math is syntax, not a function call. After the chisel, the substrate is no longer the creator's problem. The remainder surfaces: what's left when you chisel away substrate obligation? The creative coding practice itself — the act of moving from intuition to image, from description to pixel — which never needed to know what chip it ran on, but was forced to know, for twenty years, by every tool it had. Dither is the first language to let that practice exist without inheriting substrate identity.

"Dither" has a specific technical meaning in image processing: dithering takes the quantization error — the gap between the color you want and the color the hardware can actually store — and distributes it spatially across neighboring pixels, simulating tonal depth beyond the hardware's bit limit. The error is not discarded; it's redistributed. The image looks richer than its actual color depth. A creative coding language named Dither performs the same operation: it takes what fell into the gap between CPU and GPU practice — the creative potential lost to substrate bifurcation — and redistributes it as productive creative capacity. The name encodes a remainder-operation, not a metaphor. This matters now more than later: v0.0.3 was released in April 2026 and is still pre-release. Fifty GitHub stars. In the issue tracker, someone is still asking how the GPU backend actually works. The logic is still growing, still unassigned to any existing category. Once there are enough users, once a paper places it in a lineage, once the p5.js or Processing communities absorb it, it will have moved from remainder to construct. Until then, it is the held posture of a chisel-construct move still in progress.

Lingdong Huang's other projects — a drawing language built from lambda calculus, a version control system for visual sketches, a gestural computation experiment — all circle the same question: what lives in the gap between creative intuition and the tools that mediate it? Dither is that question landing at the level of a programming language. The gap is its remainder.

Han Qin (秦汉) · @hqinjarsy · SAE Judgment & Aesthetics