Saving, Loading, and Sharing Workflows

Workflows Live in Two Places

Every time ComfyUI generates an image, the complete workflow that produced it — every node, every setting, every connection — is automatically embedded in the metadata of the output PNG file. You do not need to save anything manually for this to happen. Open the PNG in an image viewer and the metadata is invisible, but drag that same PNG back onto the ComfyUI canvas and the entire workflow is restored exactly as it was.

This is one of ComfyUI's most powerful features. It means that when someone posts a ComfyUI-generated image online and shares the PNG file, they are implicitly sharing the workflow that produced it. You can reproduce not just the final image but the exact process.

Workflows can also be saved and loaded as JSON files — plain text files that describe the node graph. JSON files are smaller than PNGs and easier to version-control, share via text message, or store without carrying the output image.

Saving a Workflow as JSON

To save the current workflow as a JSON file:

• Click the workflow name in the top bar to open the file menu, or press Ctrl+S (Cmd+S on Mac).
• Select Save to save to the internal Workflows library (accessible via the Workflows panel in the left sidebar), or Export to download the JSON file to your local machine.

Saved workflows appear in the Workflows panel in the left sidebar. Click any saved workflow to load it on the canvas, replacing the current one. You will be prompted to save unsaved changes before the switch happens.

Exported JSON files can be shared directly with other ComfyUI users. Anyone with the same custom nodes installed can import and run the workflow immediately. Nodes that are missing will show as red boxes — ComfyUI-Manager can install the missing ones automatically.

Loading a Workflow: Three Methods

There are three ways to load a workflow onto the canvas:

• Drag a PNG: Drag any ComfyUI-generated PNG file directly onto the canvas. If the PNG has embedded workflow metadata, the workflow loads instantly. This is the most common method for working with community workflows shared as image files.
• Import JSON: Use the File menu (top bar) → Import or drag a .json file onto the canvas. The workflow loads from the JSON file.
• Workflows sidebar: Open the Workflows panel in the left sidebar, click a saved workflow. Loads the most recently saved version of that workflow.

If you drag a PNG that was not generated by ComfyUI (or was generated by ComfyUI but had its metadata stripped by an image editor or social media platform), nothing will load — you will just see the image placed in an Image node.

Finding Community Workflows

Two main sources for community-created workflows:

• OpenArt (openart.ai/workflows): Searchable library of ComfyUI workflows, each with a preview image and a list of required custom nodes. Many are downloadable as JSON files directly. Search by task type (portrait, video, upscale, etc.).
• Civitai (civitai.com): The primary model hub also hosts workflow examples. Many model pages include recommended workflows in their description, often with embedded PNG files or JSON downloads. Filter by "Workflows" in the content type filter.

When loading a community workflow for the first time:

• Download the JSON or PNG file.
• Drag it onto the ComfyUI canvas. Red boxes indicate missing custom nodes.
• Open ComfyUI-Manager and run Install Missing Custom Nodes. It detects which nodes are missing and installs them in batch.
• Restart ComfyUI after installing nodes.
• Check the model dropdown in each loader node — replace any placeholder model names with ones you have downloaded.

This process (load → install missing nodes → fill in models → run) becomes fast with practice and unlocks access to the entire ComfyUI community's workflow library.

What Gets Saved and What Doesn't

A saved workflow (JSON or embedded PNG) includes: every node and its position on the canvas, all widget values (settings, prompts, seeds), and all connections between nodes. It does not include: model files themselves, generated output images, or any files in the custom_nodes folder. The workflow is a recipe, not the ingredients.

This means you need the same models installed locally to reproduce a workflow exactly. A workflow that specifies dreamshaper_8.safetensors in the Load Checkpoint node will show an error if you have not downloaded that model. Replace it with a model you have, and the workflow will run — though the output will differ because the model is different.