Invoke

What Invoke is

Invoke is a professional image-generation engine rather than a hosted service. It is a creative toolkit for producing images with diffusion models, and it takes the position that the people using it should own both the pipeline and the output. The project is open source under the Apache 2.0 licence and runs on your own hardware, so prompts, models and finished images never leave your machine.

There is a naming trap worth clearing up first. The address recorded for this listing did not serve a readable product when checked, while the project itself lives at its own published address, where it is distributed as a community edition for Windows, macOS and Linux. The name also collides with unrelated products, so be precise about which address you mean before downloading anything.

The creative engine

The workflow centres on a layered canvas rather than a single prompt box. You can paint, mask, sketch and edit on separate layers, which makes it practical to iterate on one part of an image without regenerating the rest, and it is closer to how a designer already works than the usual one-shot generation model.

Alongside the canvas sits a node-based graph, where you assemble a pipeline visually so that a complex, repeatable process can be built once and reused. Parameters can be exposed as a simplified interface for other people to run without touching the graph itself. Model management covers the common bases: base checkpoints, LoRA adapters, textual inversion embeddings and ControlNet models, with ControlNet guidance by depth, edge and pose for tighter control over composition. Iteration tools include batch generation, prompt wildcards and high-resolution upscaling, which together are what make the difference between a demo and a working studio tool.

For anyone training their own models, a dedicated interface handles textual inversion and LoRA training, and third-party GPU hosts are available if you would rather not run everything locally.

Who it is for

Artists, designers and creative teams with a production workflow, who want control over the pipeline rather than a subscription to somebody else's, and who are comfortable installing software. It also suits people who need to keep client material off third-party servers, and anyone who wants to train a model on their own material.

What to keep in mind

Self-hosting is the trade. You need a capable GPU and some appetite for setup, and you are the one managing updates, models and disk space. The vendor-run hosted version has been discontinued — the founding team moved to Adobe — so if you want cloud convenience, that comes from third-party GPU hosts rather than the project itself.

Platform support is uneven: the macOS build is Apple Silicon only, so older Intel machines are out. And note the scope: this generates images, not video, so plan around that rather than assuming it will grow into video work.