Keep a ComfyUI workflow you can reuse.

Keep the JSON and its dependencies.

A workflow file is only part of a reproducible setup. Record the ComfyUI version, required custom nodes, model filenames or revisions, and relevant input files alongside it. A missing node or a different model can change whether the graph runs.

Export the workflow from ComfyUI, then import the JSON text into a private setup in RMA. Review its title and notes before saving. RMA preserves the text for copying or downloading; it does not execute the workflow or inspect installed nodes.

Label the format.

ComfyUI’s editor save format and API format serve different purposes. The official API guide describes a separate execution graph for programmatic requests. Keep the original editor workflow when you need to edit the graph, and label an API export so you do not confuse the two.

A JSON file passing syntax validation does not establish compatibility. Open or submit it using the intended ComfyUI path, then confirm that the expected output is produced.

Make the notes useful to your future self.

Record one small example that you have actually run: image dimensions, seed if relevant, model choice, required nodes, and any changes you made. Describe the result rather than promising identical output on different hardware or versions.

# Workflow notes
Workflow name:
Format: editor save / API
ComfyUI version:
Custom nodes and versions:
Models and revisions:
Required input files:
Test dimensions / settings:
Observed output:
Known limits:
Original source and license:

Reopen, copy, download.

Search your saved setups by name or description. Open the record to edit the notes, copy its JSON, or download it for your local tool. Archive experiments you no longer use and restore them when needed.

The current workspace accepts text imports up to 40,000 characters, subject to the request-size limit. For larger workflows, keep the file in your own repository and save its URL with a concise setup note. Do not paste secrets or private access links.