Pinned
Guides Moving a ComfyUI setup off a network volume
Your setup works on a network volume, but the GPUs you want are in another region or on another cloud. Pack it once, restore it where the GPUs are.
Read it →Blog
What we shipped, how to move a setup, and what we learned along the way.
Your setup works on a network volume, but the GPUs you want are in another region or on another cloud. Pack it once, restore it where the GPUs are.
Read it →Send someone a ComfyUI workflow and they get the graph. The models are the hard part. We handed one to another account and ran it on their machine.
Field reportsFace swap is one of the setups people most often lose when a rented machine goes away. Thirteen model files you never placed, a GPU build of onnxruntime, and a licence question.
Field reportsA video setup is big. We restored one on four cards: four to ten minutes, and the longest wait wasn't the card or the download.
Field reportsOne setup, built once on an A40, restored on an A5000, a 3090, a 4090 and a 5090. The card that renders fastest isn't the one that costs you the most time, and the pictures aren't all equally close.
ReleasesA What's next summary after every restore, a new renest start command, a pre-flight check that touches the GPU, and plugins that fail to load named by name.
MechanismsA folder of verified files is not a working environment. Three host-side ways a restore can pass every checksum and still fail — and what the pre-check refuses now.
Self-reportsEvery file matched, the image regenerated — on the machine that packed it. The next day a clean machine couldn't install its dependencies. What went wrong and what the tool does now.
Self-reportsThe driver floor in our pre-check was copied from a real vendor table — for installing the CUDA toolkit, which we never do. It took three corrections and one paid-for measurement to get one number honest.
Field reportsFive hosts with the same GPU model regenerated one image. The results fell into exactly two byte-identical files — and the CPU vendor, not the driver, sorted them.
Field reportsSame GPU model: the regenerated video is literally the same file. Different model — even same architecture: a uniform ~0.94 plateau. And the 0.9814 mystery is solved: it was the CPU vendor.
Field reportsA workflow with 3 custom nodes and a LoRA is captured and revived intact — including a node that was already broken — while new data forces honest revisions to Vol. 1.
Field reportsJuly 2026: one ComfyUI + SDXL environment packed, the machine destroyed, and the nest restored 26 times across 2 clouds, 7 datacenters and 6 GPU types, every file verified. The pass line used here has since been replaced by a stricter one; the current record is on the Proof page.
Follow along
No schedule we can't keep.