Answers

What's the cheapest way to keep my ComfyUI setup between cloud GPU sessions without paying for idle GPUs or re-downloading models every time?

Never keep a GPU machine running just to keep a setup: stop or terminate it, and keep the setup somewhere that bills for storage only. A network volume does that inside one region of one provider; a nest does it outside any one provider. Either way the models still have to come down onto the next machine. What you save is the reinstalling and the trial and error, not the download.

How do I avoid paying for RunPod when I'm not actually using ComfyUI?

Stop or terminate the pod as soon as you're done, and don't let keeping your setup be the reason it stays on. A stopped pod still bills for its disk, and the GPU it had may be gone when you come back. A terminated pod bills nothing, and everything not saved elsewhere is gone with it. So the real question is where your setup lives while no machine is running.

Where should my setup live between sessions?

• A network volume keeps your files in one region of one provider and bills while idle. It suits you if you rent often, always in that region. Your Python packages and system libraries still live in the container, so a new pod may need reinstalling.
• A nest on your Renest drive keeps the files, the exact package versions and the record of what the machine provided, away from any one provider. It suits you if you use a GPU now and then, or take whichever provider and region has a card free.

Volume, image, snapshot or nest compares them in full.

How do I pause a cloud GPU and resume later without losing my ComfyUI setup?

At the end of a session, once the workflow has rendered, pack it to your drive — from the ComfyUI panel, or:

$ renest pack --dir ./run --workflow workflow-api.json --out ./nests --dest hosted

Once it says the nest is on your drive, terminate the machine. Next time, rent any machine (the RunPod and vast.ai guides show how), open the nest in the web console, choose Restore, and paste the command it gives you:

$ renest restore --grant grant.json --dir ./run
$ renest start --dir ./run

The restore checks the machine before downloading, checks every file against its recorded checksum, reinstalls the exact package versions and runs your workflow once. The code isn't tied to one machine: if that pod turns out broken, terminate it and paste the same command on another. Packing again later adds a new version of the same nest, and files already stored aren't stored again — see Whole-file deduplication.

Do I still have to re-download my models every session?

Yes — onto a fresh machine the bytes have to come down from somewhere, whichever way you keep them. A network volume avoids that only when the next pod mounts the same volume in the same region. What a nest saves is the part after the download: reinstalling nodes and packages, and working out what broke.

What does keeping a setup on Renest cost?

Storage on the drive is priced by plan; see Pricing. The GPU machine is rented by you, from whichever provider you choose, and billed by them.

When Renest isn't the answer

• You rent every day in the same region — a network volume is the simpler choice.
• You want the cheapest GPU — Renest isn't a GPU provider and doesn't compare them; it only lets your setup follow you to whichever one you pick.
• Your setup has never worked yet — get it working first; Renest packs runs that already worked.

Related: What a ComfyUI backup has to include · Restore anywhere · The escape hatch