A ComfyUI workflow JSON file describes the workflow graph: nodes, connections, layout, and related configuration. Read it as a recipe for a graph—not as a bundle containing every model and custom node needed to reproduce the output.
The short answer
A ComfyUI workflow JSON file stores a node graph in a human-readable data format. It tells ComfyUI which nodes exist, how outputs connect to inputs, and how the graph is arranged. It may reference checkpoint names, LoRAs, VAEs, custom nodes, prompts, seeds, dimensions, and sampler settings.
The JSON normally does not contain the model weights or custom-node implementation. Loading the graph can therefore produce missing-node or missing-model warnings even when the file itself is valid. Reproduction depends on matching those external dependencies and the relevant software versions.
Start with the graph, not the raw punctuation
Open the workflow in a compatible ComfyUI version and follow the main path from model loading through positive and negative conditioning, sampling, decoding, and saving. Branches may add ControlNet, upscaling, masking, video, or other stages. The visual graph is usually easier to audit than reading thousands of braces and numeric node IDs.
Use the raw JSON when you need to compare versions, inspect a suspicious reference, repair a merge conflict, or understand which values are stored. Do not hand-edit node IDs or links unless you understand how the schema connects them.
Build a dependency inventory before running it
List every checkpoint, LoRA, VAE, embedding, ControlNet model, and custom node the graph expects. Record filenames and versions rather than downloading the first search result with a similar name. Two files with the same everyday label can target different architectures.
Treat custom nodes as code. Review their repository, maintenance state, installation instructions, dependencies, and permissions before installing them. A shared JSON graph can reference a node, but it does not establish that the node is safe or appropriate for your machine.
What to include when sharing a workflow
Share the JSON plus a short dependency manifest: ComfyUI version, custom-node names and repository links, model names and versions, required trigger phrases, expected input types, and the output the workflow is designed to create. State whether any commercial or restricted model license applies.
Remove private prompts, local paths, API endpoints, account identifiers, access tokens, or embedded secrets before publishing. Search the JSON as text and review the visible nodes. A workflow is portable configuration, so it can also carry information you did not intend to disclose.
Workflow JSON versus image metadata
ComfyUI can store a workflow in the metadata of supported generated images, allowing the graph to be recovered from the output. A separate JSON file is useful for formats without suitable metadata, version control, collaboration, and archiving independent of one image.
Before relying on image metadata as a backup, export the workflow JSON explicitly and test reloading it. Social networks, image editors, optimization tools, and format conversion can strip metadata without changing the visible picture.
PhoneDiffusion availability and article scope
PhoneDiffusion does not run ComfyUI or import its workflow JSON files, custom nodes, desktop checkpoints, or arbitrary LoRAs. PhoneDiffusion uses curated on-device model packs and its own supported generation and editing controls.
A manual hybrid workflow is still possible: preserve the prompt, seed, image, and settings you can identify, then rebuild the intended stage in the other tool. Different models and runtimes may not reproduce identical pixels.
Field notes
- Workflow JSON stores the graph and configuration, not every model or custom-node dependency.
- Audit models and custom nodes before running an unfamiliar workflow.
- Share a dependency manifest and remove private paths, prompts, endpoints, and secrets.
Try it on your phone
- Open the workflow in a compatible ComfyUI version and identify its main generation path.
- Inventory every model file, custom node, input, and expected output before queueing it.
- Verify each dependency’s source, architecture, version, and license.
- Remove private data and export a clean JSON plus a short dependency manifest.
- Reload the shared copy in a separate test environment before calling it portable.
Sources and verification
Primary and maintainer documentation checked for this guide. Product-specific details were reviewed against the current PhoneDiffusion scope.
- ComfyUI — workflow concepts and savingOfficial explanation of node graphs, workflow metadata in generated images, and portable JSON workflow files.
- ComfyUI — Workflow JSON specificationOfficial schema reference for the data stored in a ComfyUI workflow JSON file.
- ComfyUI — official documentationOfficial overview of ComfyUI as a node-based interface and inference engine, including local and cloud operation.