Skip to main content

Documentation

Documentation/ComfyUI Workflows

ComfyUI Workflows

ModelPilot can inspect ComfyUI workflow JSON, identify recognized models and custom nodes, recommend a GPU, and prepare supported files for a dedicated deployment. Compatibility depends on the exact graph; arbitrary workflows are not guaranteed to run unchanged.

Safest path: start with a verified workflow

A verified workflow page includes the downloadable JSON, exact model filenames, expected directories, preview evidence, GPU guidance, and known limitations. You may download and run the JSON yourself without deploying it on ModelPilot.

Check your own workflow

  1. Export the workflow in ComfyUI frontend or API JSON format.
  2. Open the workflow checker and paste or drop the JSON.
  3. Review detected built-in nodes, custom nodes, models, unknown items, warnings, and the recommended GPU.
  4. Correct unknown filenames or node sources before deploying. A downloadable lockfile records what the checker recognized.
  5. Continue to deployment only after the requirements match the graph you intended to run.

Workflow checker privacy

The web checker sends the workflow JSON securely to ModelPilot for server-side analysis and does not store it. It does not upload the image, video, model-weight, or output files referenced by the graph. Remove secrets from node fields before submitting any workflow.

What the checker can and cannot prove

It can identify

  • Known built-in and custom node types
  • Recognized model filenames and download mappings
  • Likely GPU capacity and dependency warnings
  • Unsupported or unknown requirements

It cannot guarantee

  • That every custom-node version is compatible
  • That gated or removed files can be downloaded
  • That the graph fits memory at every resolution or batch size
  • That the workflow produces the desired output

Prepare a custom workflow for deployment

  • Run the workflow successfully in ComfyUI first.
  • Use exact relative model filenames, not paths from one computer.
  • Prefer public, versioned custom-node repositories.
  • Confirm each model file is publicly downloadable or provide the required access through the deployment settings.
  • Start with conservative resolution, frame count, and batch size.
  • Treat the recommended GPU as a starting estimate and review every warning before paying to deploy.

If deployment fails

Check the first model-download, custom-node import, or out-of-memory error in the deployment logs. Do not repeatedly redeploy an unchanged workflow. Follow the troubleshooting guide and include the deployment ID if you contact support.