Dimensions AI

Stable Diffusion, given a home.

In early 2023, generative image models lived in notebooks and command lines. Dimensions AI was an attempt to give Stable Diffusion the thing it lacked — a place to live. A native desktop application that wrapped the full diffusion pipeline in an interface you could actually create in, with nothing leaving your machine.

It was equal parts art tool and research bench: a way to push the model past its defaults with custom training, specialized architectures for different art styles, and an inference path tuned for real-time generation.

An Electron shell over a Python core.

The interface is built in Electron — a fast, familiar desktop surface — while the heavy lifting happens in a bundled Python backend running the diffusion stack. The two talk over a local bridge, so the UI stays responsive while the GPU works.

On top of stock Stable Diffusion sit two layers of my own: a custom training pipeline for fine-tuning the model on new styles, and a set of optimization techniques — memory-efficient attention, half-precision inference and scheduler tuning — that pull generation close to real time.

Depth, in cyan.

Dimensions takes its name from latent space — the high-dimensional field a diffusion model samples to find an image. The identity leans into that idea: a cool, electric cyan against near-black, the colour of a signal cutting through depth.