Local image generation runtimes
An image generation app, sometimes called a runtime, is the program that turns a prompt into an image on your own machine. It sits between you and the model weights you downloaded, and it does the unglamorous work: loading the model, running the diffusion sampling loop, managing memory, applying extras like upscalers and LoRA adapters, and writing the finished file to disk. This page is part of the image generation section, and it covers the three apps most people end up choosing between.
The models themselves are separate from the app. A checkpoint you download for one app will generally load in the others, because they all read the same weight formats and run the same sampling algorithms underneath. Picking an app is therefore a low-risk decision. You are choosing an interface and an ecosystem, not locking your model collection to one tool.
Why the choice matters
Three things actually differ between them, and they are worth thinking about before you install anything.
- Workflow style. Some apps hand you a form with prompt boxes and sliders. Some hand you a graph where every step is a node you wire up yourself. Some build the interaction around a canvas you paint on. The same image is reachable from all three, but the effort to get there is very different depending on how you like to work.
- Extension ecosystem. New techniques usually show up first as third-party extensions or custom nodes. An app with a large community gets support for new model families and control methods quickly. A smaller, more curated one gets them later but with fewer broken installs.
- Hardware backends. All three run on NVIDIA GPUs through CUDA, and all three have working paths for Apple Silicon and AMD cards. Support quality is not identical across those paths, and neither is the amount of tuning you will need to do. See the GPU pages for what each class of hardware can hold.
The three runtimes
- ComfyUI is a node-graph workflow tool. You build the generation pipeline explicitly, connect stages together, and save the graph as a reusable workflow.
- Stable Diffusion WebUI is the form-based interface many people start with. Prompt, settings, generate, with tabs for the common extras.
- InvokeAI is a canvas-oriented studio application built around iterative editing of a single image rather than batch generation.
If you have no strong preference yet, start with the form-based interface, learn what the settings do, and move to a node graph when you find yourself wanting steps it cannot express. There is no cost to running more than one against the same model folder, and organising your model files sensibly makes that easy.