Comfy Cloud or local?
The trade-off between running models on Comfy Cloud (pay per render, nothing to install) and your own ComfyUI (free to run, but you need the hardware).
Models can run two ways, and you can use either or both. The right choice depends on how heavily you generate, what hardware you have, and whether you would rather pay per render or invest in a capable graphics card.
Comfy Cloud
The model runs on rented hardware.
- Nothing to install -- no downloads, no ComfyUI to manage.
- Always the full-size model -- the cloud hardware has the memory, so you get full quality.
- No graphics card needed -- your own computer only sends the job and receives the result.
- You pay for each render -- generating costs credits, so heavy use adds up.
Comfy Cloud is the easiest way to start, the simplest path for demanding jobs like video, and the answer when a model is too big for your card.
Local ComfyUI
The model runs on your own computer.
- Free to run once set up -- no per-render charge; you are using hardware you already own.
- Private and under your control -- files and generations stay on your machine.
- But you download the files -- models are multiple gigabytes each, and you need enough graphics memory to run them (see Model versions and quantization).
- And you run ComfyUI yourself -- installing and keeping it working is on you.
Local is worth it when you generate a lot, like to tinker, want to work offline, or would rather not pay per render.
A rough cost picture
Comfy Cloud charges credits per render, so its cost scales with how much you generate. Local generation has no per-render charge -- the cost is the electricity, a one-time download per model, and a graphics card capable enough to run what you want. If you generate occasionally, the cloud is usually cheaper overall; if you generate constantly, a capable local setup pays for itself.
Using both
You do not have to choose once and for all. You can enable Comfy Cloud and local ComfyUI together and pick per job -- for example, drafting locally and running the final, heaviest step on the cloud.
Setting each up
Both live under Settings, in the ComfyUI section:
- Comfy Cloud -- add your Comfy Cloud API key.
- Local ComfyUI -- point the platform at your running ComfyUI, and connect your models folder so it can see which files you already have and install missing ones for you.
The AI Models page then guides the rest: it asks which of these you want to use, and shows the files any models you turn on will need.