Execute ComfyUI image and video workflows on RunPod serverless GPUs via a JSON-output CLI.
npx skills add https://github.com/Hearmeman24/ComfyGen --skill comfy-gen-hearmeman24
Please help me install this Agent Skill. Skill: comfy-gen Source: https://github.com/Hearmeman24/ComfyGen Command: npx skills add https://github.com/Hearmeman24/ComfyGen --skill comfy-gen-hearmeman24
--override 7.seed=42) and input file mapping (--input 193=photo.jpg), polls with real-time progress, and returns pre-signed S3 output URLs with model hashes. - Infrastructure & Model Management: Sets up RunPod endpoints via comfy-gen init, lists installed models, queries available samplers/schedulers, and cancels or checks job status. - Use Case: A user shares a Wan 2.2 video workflow. The Skill identifies five required models, locates them on HuggingFace, batch-downloads them to the volume, then submits the workflow with a 900-second timeout and returns the video URL. ## Quick Start Ask the agent to analyze your ComfyUI workflow JSON, download any missing models to the RunPod volume, and submit it for generation.High-intent search queries and answers about installing and using this skill.
Run comfy-gen submit workflow.json with a ComfyUI API-format workflow file. The CLI uploads input files to S3, submits the job to your RunPod endpoint, polls for completion, and returns a JSON result with the output URL.
Use comfy-gen download with either a CivitAI version ID or a direct HuggingFace URL, plus a --dest subfolder like loras or checkpoints. Files land at /runpod-volume/ComfyUI/models/<dest>/ via a serverless job.
Workflows must be in ComfyUI API format, the node-ID-keyed JSON with class_type and inputs fields. Export it from the ComfyUI interface using Save (API Format), not the standard UI save format.
Yes, any S3-compatible storage works. Set s3_endpoint_url to your provider's endpoint (for R2 use region auto), along with access key, secret key, and bucket name in the config.
Video generation often exceeds the default 600-second timeout. Pass --timeout 900 or higher when submitting, and note the RunPod endpoint's executionTimeoutMs may also need increasing for very large model downloads.
Use the repeatable --override flag with node_id.param=value syntax, such as --override 7.seed=42 --override 7.denoise=0.8. Numeric values are auto-coerced to integers or floats.
huivrotikiQueue ComfyUI prompts and manage workflows on RunPod via OpenClaw agents.
miles990Develop content management systems with TypeScript content modeling and Tiptap editing.
Nitrogen216Verifies every numeric claim in a paper against raw result files using a zero-context cross-model reviewer.
williamsforealRoutes image-generation requests between fal.ai and local ComfyUI based on cost and latency.