RunComfy for n8n

Run ComfyUI workflows, connect an AI image generator or AI video generator to your automation, and track LoRA training with RunComfy.

Search AI image models and AI video models, inspect current schemas and prices, submit a generation, and retrieve image, video or checkpoint URLs. RunComfy handles GPU execution. This native community node connects directly to RunComfy's REST APIs.

The RunComfy models directory includes FLUX 2, Seedance 2.5, Wan 3.0, Wan 2.2/2.7, LTX 2.5 and Seedream 5.0. Availability, exact IDs and inputs come from the live API. Use Model → Get before configuring generation; models and pricing can change.

RunComfy · API documentation · Models · Support

Installation

Package: n8n-nodes-runcomfy.

After the npm release is available, install through Settings → Community Nodes → Install on self-hosted n8n. Enter the package name and follow the prompts. See n8n's installation guide.

n8n Cloud installation requires n8n verification. A public repository or npm release does not imply verification. See the publishing checklist. To use source locally, see Development.

Credentials

1. Sign in to your RunComfy Profile and create an API token.
2. In n8n, create a RunComfy API credential.
3. Paste the token into API Token, without Bearer , and test the connection.

n8n stores the token in its credential system. The connection test reads the account balance. Requests use fixed RunComfy API hosts and do not follow redirects. The node does not read files, environment variables or local token stores. The REST APIs use a Bearer API token; the ChatGPT/MCP connection is separate.

Operations

| Resource | Operations | | -------------------- | ---------------------------------------------------------- | | Model | Get Many, Get, Generate Image, Generate Video, Run | | Model Request | Get Status, Get Result, Cancel | | ComfyUI Workflow | Get Many, Get, Run, Get Status, Get Result, Cancel Request | | Training Job | Get Status, Get Result | | Account | Get Balance |

Generate Image and Generate Video check the selected model's category and required fields before submission. Run accepts any supported category, including compatible LoRA inference. All use the selected model's actual input schema.

Training Job reads existing jobs created in RunComfy Trainer. This version does not create datasets, upload training files, or start/resume training jobs.

Generate an image

1. Use Model → Get Many to search FLUX or Seedream.
2. Use Model → Get to inspect the chosen model's schema and current price.
3. Add Model → Generate Image, choose the model, and enter Inputs as a JSON object matching its schema. Map upstream values using n8n expressions.

For blackforestlabs/flux-1-schnell/text-to-image:

{
	"prompt": "Studio photograph of a blue ceramic mug on a white tabletop",
	"image_size": "square_hd",
	"seed": 123456791
}

Store the returned request_id and model_id. Use Model Request → Get Status, then Get Result after status becomes completed.

Generate a video

Search Seedance, Wan or LTX, inspect the model with Get, then use Generate Video. For image-to-video, include the image URL and other required fields. File inputs must be accessible to RunComfy over HTTPS. Models do not all use the same duration, aspect-ratio or image fields.

Video generation continues on RunComfy while the workflow waits. The submission node returns the request ID immediately.

Run a saved ComfyUI workflow

1. Create and deploy your workflow in RunComfy.
2. Select ComfyUI Workflow → Get with Include Saved Inputs enabled.
3. Inspect the saved node IDs and input names.
4. Use Run → Override Saved Inputs to change selected inputs:

{
	"6": {
		"inputs": {
			"text": "A blue ceramic mug on a white tabletop"
		}
	}
}

6 is an example; use your actual deployment's node ID. Empty {} uses saved defaults. Complete Workflow JSON accepts a complete ComfyUI export in API format.

Use deployment_id and request_id with ComfyUI Workflow → Get Status / Get Result. Workflow calls use current /prod/v2 endpoints. Model API IDs use Model Request instead.

Waiting, results and costs

• Each input item sent to Run, Generate Image or Generate Video can incur charges. Multiple input items create multiple requests.
• Store the returned ID before waiting. The submission node does not sleep or automatically resubmit.
• Use n8n's Wait node and a bounded status-check loop, or a supported workflow webhook. Stop on terminal status and impose a maximum wait time.
• Included examples perform one status check and stop if generation is still running. Continue checking the saved ID; do not rerun submission to fetch an existing result.
• Leave Retry On Fail disabled for paid submission nodes unless you have checked whether the original request was accepted. A timeout can have an uncertain outcome; workflow-wide retries can also create duplicates.
• Get Result returns hosted URLs without downloading media. Use n8n HTTP Request or storage nodes to process outputs.
• Cancellation can return not_cancellable; it does not guarantee a running job stopped or charges were refunded.
• Get Balance is an account-wide snapshot. Warm GPU charges may settle after inference completes.

Output supports raw, simplified or selected fields. Raw is the default to preserve model-specific outputs and training artifacts. Selected output retains identifiers returned by the API.

Examples

Import these inactive workflows, assign your credential, and review configuration before running:

• Discover FLUX and Seedance models — read-only discovery.
• Generate an image and check once — one paid request and one status check.
• Run a ComfyUI workflow and check once — the request-ID pattern for a saved deployment.

No example includes credentials, private account IDs, or a live webhook address. All begin with a manual trigger.

Compatibility and support

There are no runtime dependencies. The node uses n8n credential/HTTP helpers and the Model API, Serverless API, and Trainer API.

The first release is developed against Node.js 24 and n8n-workflow 2.37.4. Older n8n compatibility is not claimed until tested. Report integration issues in GitHub Issues. For account or billing help: [email protected]. Never include tokens in issue reports.

Development

npm ci
npm run check
npm run dev

Scaffolded using the official @n8n/create-node programmatic template. Programmatic execution supports a dependent schema lookup before a paid request, catalog pagination, and distinct request lifecycles. Authentication remains delegated to n8n.

Tests cover routing, required-field checks, uncertain submissions, pagination, item pairing and output preservation. See API references and validation. The package uses the strict default n8n lint configuration.

License

MIT. The name and logo identify RunComfy. The icon embeds the existing official RunComfy favicon without alteration.