runpod-workers/worker-comfyui

What it solves

It enables users to run ComfyUI workflows as serverless API endpoints on the RunPod platform. This removes the need to manage a persistent GPU server for image generation, allowing developers to trigger complex ComfyUI workflows via API calls and receive results programmatically.

How it works

The project provides specialized Docker images that bundle ComfyUI with various model options (such as FLUX.1, SDXL, and SD3). Once deployed as a RunPod serverless endpoint, the worker accepts JSON requests containing a ComfyUI "API format" workflow. It processes the request, generates the images, and returns them either as base64 strings or as URLs uploaded to an S3 bucket.

Who it’s for

Developers and AI artists who want to integrate ComfyUI's node-based generation power into their own applications or automate image generation pipelines without managing infrastructure.

Highlights

• Pre-configured Images: Offers Docker images with built-in support for FLUX.1 (schnell and dev), Stable Diffusion XL, and Stable Diffusion 3.
• Flexible Output: Supports returning images as base64 strings for immediate use or S3 URLs for persistent storage.
• Async and Sync Support: Provides both /runsync for immediate results and /run for asynchronous processing with polling or webhooks.
• Customizability: Allows for custom models and nodes via network volumes or custom Docker builds, and supports SSH access for debugging.