ComfyUI
Node-graph interface for image models. The most capable local image generator — and the clearest example of a tool that does not belong on a NAS.
ComfyUI is included here mostly as the counter-example. The other tools on this site are chosen because they run on hardware you already own. ComfyUI is chosen because people search for it and deserve an honest answer: it wants a GPU, and a CPU-only box will disappoint you.
The reason is arithmetic. Large image models do billions of operations per image, and a vectorised GPU does that in seconds while a CPU does it in minutes. There is no quantisation trick that closes a fifty-fold gap. A NAS with an integrated GPU from 2011 does not participate in this at all.
It is still the best tool for the job it does
If you do have a card — even a modest 8 GB one — ComfyUI is the most controllable image pipeline available. Workflows are JSON files, so a look you like can be saved, versioned and re-run exactly. Compare that to re-typing a prompt into a different UI and hoping for the same seed behaviour.
Two things worth knowing before you start:
- The graph is the documentation. Most tutorials show a screenshot of nodes; you copy the layout, then read what each node does.
- Model choice matters more than node count. A well-chosen checkpoint with a simple graph beats a 60-node graph built on a bad one.
Verdict
Buy or borrow a GPU, or use a cloud endpoint. Do not install this on the NAS and conclude that local AI is slow — you would be measuring the wrong hardware.
Pros
- The most flexible image workflow there is: every step is a node you can rewire
- Most new model releases are available here first
- Runs on 4 GB of VRAM with SD 1.5-era models, if you keep the resolution down
Cons
- On CPU, a single 1024x1024 image takes minutes — this is not a "leave it on the NAS" tool
- Node graphs are a programming language wearing a GUI; the learning curve is real
- Model files are large (2–7 GB each) and multiply quickly
Alternatives worth a look
Ollama
The default way to run a local LLM. One binary, a model library, and an OpenAI-compatible endpoint on port 11434.
Open WebUI
A self-hosted ChatGPT-style interface that sits on top of Ollama or any OpenAI-compatible API. Accounts, history, RAG and document upload included.
Last verified 2026-10-17 (docs). Prices and licence terms change — check the vendor page before you buy. Spotted an error?