What it does
You assemble a chain of nodes — prompt, references, model, post-processing — and watch results land on one canvas. References, upscale, image-to-video, lip-sync, batch generation, a gallery, an LLM prompt assistant. Models are routed through a gateway with per-provider pricing, so a user sees the honest price in their currency before pressing the button.
Money and API
Two independent payment rails, webhook handling, one-time top-up tokens against double crediting, balances, transaction history, a welcome bonus. A versioned REST API with user keys and its own docs page — so the same engine serves other people's products, not only the canvas.
Two things I'd show an engineer
An incident with the irreversible loss of 237 user images — diagnosed, and the media storage scheme redesigned so it can't repeat. And a 116 KB UX audit of my own product: six parallel interface reviews with findings verified before acting, which reshaped the key flows.