Run ComfyUI as an API

RunComfy Serverless API turns your ComfyUI workflows into production-grade ComfyUI APIs with auto-scaling and no operations needed. This lets you focus on building generative AI without infrastructure worries. Unlike local ComfyUI setups that require hardware management, CUDA setup, and ongoing monitoring, RunComfy Serverless API handles deployment, scaling, and consistency in the cloud. Your ComfyUI API runs reliably on high-performance GPUs, making it easy to go from prototype to production. For more details, please read the RunComfy Serverless API documentation.

To deploy a ComfyUI workflow as a ComfyUI API service on RunComfy, start by building it in ComfyUI Cloud and saving it along with your nodes, models, and dependencies. Then, select GPU hardware, set autoscaling rules, and deploy with a few clicks. This creates a serverless ComfyUI API that scales automatically, processes requests asynchronously, and provides endpoints for inferences. You'll have a ready-to-use ComfyUI API without dealing with Docker, Kubernetes, or manual configurations, everything is reproducible and consistent.

To deploy your ComfyUI workflow as a ComfyUI API on RunComfy, start in ComfyUI Cloud where you can easily create or edit your workflow. Once it's ready, export it as a simple API JSON file and pick the parts you want to tweak during runs, like prompts or seeds—this keeps things flexible. From there, just click Cloud Save. RunComfy takes care of the rest by bundling your workflow, nodes, models, and full setup into a ready-to-use container, so you skip all the technical headaches. Finally, deploy it by selecting your preferred GPU and basic scaling options. You'll instantly get a unique deployment ID to connect your ComfyUI API to your apps or projects. The whole thing is designed to be quick and hassle-free, letting you focus on your creative ideas while getting a scalable ComfyUI API without any DevOps work. For more details, check RunComfy Serverless API - Quickstart documentation.

For the latest version of ComfyUI, open the ComfyUI interface, locate the Workflow menu in the upper-left corner, and select "Export (API)" from the options. This will generate a JSON file that includes all your nodes, inputs, default values, and connections. For older versions, you need to enable dev mode in the settings (click the gear icon next to Queue Size or in the menu box, then check the "Enable Dev mode Options" box), which will make the "Save (API Format)" button appear in the menu.

RunComfy offers a range of high-performance GPUs for your ComfyUI API deployments, with VRAM from 16GB for basic workflows to 141GB for intensive models. To choose the right one for your ComfyUI API workflow, consider your model's size and memory needs, start with around 48GB (like X-Large or X-Large Plus) for most typical tasks to ensure smooth performance, then scale up or down based on testing. Monitor usage in the dashboard to optimize. For full details, visit the RunComfy Pricing page.

Yes, you can easily include custom nodes, models, or dependencies in your deployed ComfyUI API. Simply add them when saving your workflow in ComfyUI Cloud, such as custom nodes, models, or specific libraries, and they'll be bundled into the container. RunComfy automatically recreates your exact environment for consistent, reliable results every time. No extra setup is required after deployment, so you can build advanced ComfyUI APIs that fit your specific needs.

Yes, RunComfy's 200+ templates let you deploy a ComfyUI API quickly, providing workflows corresponding to the latest models. Browse community workflows, fork one, tweak nodes or parameters, and save it as your own. Then deploy it as a customized ComfyUI API. All your changes stay private.

After deploying your ComfyUI API, you have endpoints for key actions: POST to queue inferences, GET to check job status or progress, GET to retrieve results like images or videos, and POST to cancel jobs. Use your deployment_id in HTTP/REST requests, with API keys for security. This asynchronous design keeps your ComfyUI API efficient, so you can track jobs easily. For full details, visit the RunComfy Serverless API - Async Queue Endpoints documentation.

Yes, you can easily integrate the ComfyUI API with your existing tech stack. It uses simple HTTP/REST calls and JSON data, so it works with common tools like curl, Python, or JavaScript. Check the Quickstart for ready-to-use code snippets to get started fast.

Auto-scaling for your ComfyUI API increases instances during busy times and scales to zero when idle, keeping things efficient. You can set min/max instances, queue sizes, and keep-warm times to fine-tune latency and costs. You're only charged for active GPU time, with no fees for downtime. This flexible control helps you run a cost-effective ComfyUI API that matches your traffic patterns.

You can monitor your ComfyUI API with a real-time dashboard that shows request counts, queue times, cold starts, execution speeds, and usage patterns. You can also review billing data in the dashboard to track and optimize costs based on GPU time. Use these insights to adjust GPUs, scaling rules. This helps you keep your ComfyUI API running smoothly, fix issues fast, and manage expenses effectively.

To update your ComfyUI workflow without downtime, save changes as a new version under the same name, this bundles the updates into a fresh container while keeping your live ComfyUI API running on the current version. When ready, edit the deployment to switch to the new version, which rolls out gradually: existing jobs complete on the old one, and new requests use the update. Roll back anytime by selecting a previous version. This ensures your ComfyUI API stays stable and available. For more details, refer to RunComfy Serverless API - Workflow Versions and RunComfy Serverless API - Edit a Deployment.

Your workflows run on dedicated, isolated GPUs, which guarantees complete resource separation so that no processes or memory are ever shared with other users. This ensures that your computation environment remains private and independent, providing both stability and security. Each ComfyUI execution environment, including the operating system, Python runtime, ComfyUI core, workflow definitions, models, and custom nodes, is encapsulated in its own secure cloud container. These containers are persistent, allowing your entire setup to be reliably reproduced across sessions while remaining fully private to you. Access to these environments is strictly controlled: only you can manage or expose your containerized setup, and no third party, including RunComfy, can access it unless you explicitly choose to share.

Most ComfyUI workflows run smoothly with the ComfyUI API. However, very large models may require GPUs with higher VRAM to avoid memory-related issues. The number of concurrent jobs you can run depends on your scaling configuration, and queue limits can be adjusted to fit your workload. For high-volume or specialized needs, enterprise support is available, please reach out to us at hi@runcomfy.com.

Billing for the ComfyUI API follows a pay-per-use model. You are only charged for the exact number of seconds your GPU is actively running, giving you full cost efficiency and flexibility. For more details, please see the RunComfy Serverless API - Billing documentation.

If you encounter issues while using the ComfyUI API, we recommend first checking the official documentation RunComfy Serverless API – Error Handling, which covers common error codes and troubleshooting steps. If the problem persists or you need additional assistance, you can always contact us at hi@runcomfy.com.

Yes, we provide solutions tailored for enterprises and teams. For more details and customized support, please contact us directly at hi@runcomfy.com.