Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > ComfyUI-Cloud-Offload

ComfyUI Extension: ComfyUI-Cloud-Offload

Repo Name

ComfyUI-Cloud-Offload

Author
jethac (Account age: 0 days)
Nodes
View all nodes(1)
Latest Updated
2026-11-03
Github Stars
0.0K

How to Install ComfyUI-Cloud-Offload

Install this extension via the ComfyUI Manager by searching for ComfyUI-Cloud-Offload
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter ComfyUI-Cloud-Offload in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

ComfyUI-Cloud-Offload Description

ComfyUI-Cloud-Offload runs selected ComfyUI workflow nodes on rented cloud GPUs while keeping them editable locally, with cost estimates, live progress, previews, and results returned to the workflow.

ComfyUI-Cloud-Offload Introduction

ComfyUI-Cloud-Offload lets you run selected parts of a ComfyUI workflow on a rented cloud GPU. You draw a visible Cloud Offload box around the nodes you want to run remotely. Those nodes remain visible and editable on your canvas, while progress, previews, and errors appear as the remote work runs.

This can help when a workflow needs more GPU memory or computing power than your local machine has. For example, you might keep image loading and saving on your computer while sending a demanding generation step to a cloud GPU:

[Load Image] → [Cloud Offload box: KSampler → VAE Decode] → [Save Image]

The extension connects ComfyUI to a separate Cloud Offload coordinator. The coordinator manages cloud providers, queues, rented workers, and remote execution. RunPod is the default provider, and Vast.ai is an alternative. Studio fleets and pooled compute are designed but are on the roadmap, not described as currently available features.

How ComfyUI-Cloud-Offload Works

Think of the box as a temporary handoff point in your workflow. ComfyUI prepares the selected nodes and the data they need, sends that work to a remote runner, then brings the results back so the rest of your local workflow can continue.

  1. Choose the work. Select one or more nodes and use Cloud Offload selection in ComfyUI's selection toolbox. A box appears around them.
  2. Prepare the handoff. When you queue the workflow, the extension checks the connections crossing the box boundary and prepares the required data and files.
  3. Review the rental. A free preflight recommends a GPU and shows estimated cost and time before paid work starts.
  4. Watch the run. The boxed nodes remain on your canvas. You can see node-level progress, completed or cached nodes, the box's percentage, and live previews.
  5. Receive the results. Images and other supported outputs return to your local workflow.

For this handoff to work, the remote runner must have every custom node and model used inside the box. The default runner is not tied to a particular image model; it runs a pinned ComfyUI environment with the required partition components.

The box boundary also matters. Images, masks, latents, conditioning, audio, tensors, JSON-compatible values, byte buffers, and supported file-backed mesh or 3D artifacts can cross it. Live objects such as MODEL, CLIP, VAE, control nets, and samplers cannot. If a model loader feeds a node inside the box, move the loader inside the box too. The extension rejects an unsupported boundary before provisioning a paid runner.

For the underlying handoff format, see PARTITION_PROTOCOL.md.

ComfyUI-Cloud-Offload Features

Visible Cloud Offload box

The box keeps your remote nodes expanded and editable instead of replacing them with an opaque single node. During execution, it highlights the current node, marks completed or cached nodes, displays overall progress in its title, and shows live previews.

The box owns settings for provider, GPU type, minimum VRAM, timeout, and warm-runner policy. For example, a workflow that needs more GPU memory can use a higher minimum VRAM setting. A timeout determines how long the remote work may run; the warm-runner policy controls whether an already prepared runner may be reused.

GPU recommendation and rental confirmation

After the required boundary files have been uploaded, a free preflight runs before paid work is submitted. Its confirmation panel shows the recommended provider, GPU, and region; hourly price; estimated total cost and time ranges; prepared-cache coverage; and the reasoning and confidence behind the recommendation. These are estimates, with meaningful uncertainty shown where applicable.

The normal confirmation starts the job automatically after a server-controlled ten-second countdown. You can instead choose Start now, Cancel, or Choose another GPU. Opening details or changing the GPU pauses the automatic start. You can also choose Don't show this confirmation again for normal confirmations.

The Cloud Offload action-bar button lets you restore confirmation and adjust the countdown, recommendation policy, allowed regions, and hard limits for hourly price, total cost, and paid runtime. You can also adjust tolerances for material changes. Hiding normal confirmation does not remove hard limits: a material change in price, cost, capacity, region, or storage still requires a new confirmation.

Cloud Jobs panel

Open Cloud Jobs from the action bar or the fixed Cloud Jobs button to monitor work separately from the canvas. The panel can rebuild its job history after a page reload or reconnect.

A job entry shows its current stage, elapsed time, progress and estimated completion range, transfer activity, GPU and worker identity, rate and estimated spend, prepared-cache work, cancellation and billing status, and recent events. An ended job is not shown as definitively no longer billing until the coordinator receives a termination receipt for that exact rented resource.

The browser panel receives a limited job view. It does not receive your full workflow, prompts, local file paths, signed URLs, or raw coordinator job records.

Cloud Status and Cloud Workflow nodes

NodeWhat it does
Cloud StatusReturns queue counts, active-worker information, and RunPod or Vast.ai balances as JSON.
Cloud WorkflowRuns a whole API-format ComfyUI workflow through the same free preflight and rental-confirmation flow. It returns the first image and artifact-result information.

Cloud Workflow uploads input images before preflight. Completed images, meshes, and files return as artifact records rather than being embedded directly in job data. Cancellation also reaches the remote ComfyUI execution.

File and custom-data handling

The extension can prepare certain local files needed by custom nodes. A node pack may explicitly declare its assets; otherwise, the extension can discover applicable files. Local file contents are checked before preflight, and unresolved local files are uploaded before preflight runs again. On the worker, input assets go into its input folder and model assets into their appropriate model folders.

Custom values crossing the box boundary must use supported portable forms, such as dictionaries, lists, tuples, numbers, tensors, or bytes. The documented UniMate handling uses versioned dictionaries containing GLB, numeric archive, or .unimate bundle bytes; large model bundles cross the boundary in full. Returned 3d and files outputs are placed in a job-specific folder under your local ComfyUI output directory.

Troubleshooting ComfyUI-Cloud-Offload

A node or connection is rejected before the cloud run starts

  1. Check which connection crosses the edge of the Cloud Offload box.
  2. If it carries a live object such as MODEL, CLIP, or VAE, move the node that creates or loads that object inside the box.
  3. Queue the workflow again. Keep only supported portable values crossing the boundary.

This early rejection prevents an incompatible partition from starting paid cloud work.

A custom node or model is unavailable on the remote runner

  1. Identify every custom node and model used by the nodes inside the box.
  2. Check that the runner environment contains them. Having them only in your local ComfyUI installation is not enough.
  3. Retry after the runner environment matches the boxed workflow's requirements.

The extension cannot reach the coordinator

  1. Confirm that the separate Cloud Offload coordinator service is running.
  2. Check where the extension is looking: it uses CLOUD_OFFLOAD_URL first, then ~/.cloud-offload/service.json, then the localhost default http://127.0.0.1:11435.
  3. If the coordinator is remote or requires authentication, check CLOUD_OFFLOAD_TOKEN or the token path recorded in the service file.

The extension communicates with the coordinator over HTTP; it does not manage cloud-provider credentials itself.

A confirmation appears even though normal confirmations were hidden

A material change in price, estimated cost, capacity, region, or storage requires a fresh confirmation. Review the new details before starting. Hard spending and runtime limits also remain in effect when normal confirmations are hidden.

A job ends, but billing is not yet shown as stopped

Check its billing and resource status in Cloud Jobs. The panel waits for a provider termination receipt for the specific rented resource before stating that billing stopped.

Frequently asked questions

Do I need to put my entire workflow in the box?
No. You can select only the nodes that benefit from a cloud GPU, provided the data crossing the box boundary is supported.

Can I use my own custom nodes and models?
Yes, if the remote runner includes every custom node and model used inside the box. The default runner does not automatically contain everything installed locally.

Will I see what is happening on the remote GPU?
Yes. Boxed nodes report incremental progress, including the active node, completed or cached nodes, percentage, previews, and errors.

Can I review the cost before a paid run?
Yes. Free preflight provides a recommendation and estimated price, cost, and time information. The confirmation flow and your configured hard limits govern when paid work can begin.

Can I find a job after reloading ComfyUI?
Yes. The Cloud Jobs panel rebuilds from the coordinator's journal after a reload or reconnect.

Learn More about ComfyUI-Cloud-Offload

ComfyUI-Cloud-Offload Related Nodes

RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

ComfyUI-Cloud-Offload detailed guide | ComfyUI