
from comfygen15
Run ComfyUI workflows on RunPod serverless GPUs: manage endpoints, resolve and download models, submit workflows, and fetch outputs via the comfy-gen CLI.
ComfyGen provides an agentic CLI to execute ComfyUI image and video workflows on RunPod serverless GPUs. It analyzes ComfyUI workflow JSONs to resolve required models (HuggingFace/CivitAI), downloads models to a shared network volume, manages serverless endpoints, submits jobs, polls status, and returns output URLs and metadata.
Use this skill when a user wants to run ComfyUI workflows remotely, needs help finding/downloading missing models, wants to create or configure a RunPod endpoint/volume, or needs to manage or cancel generation jobs. Trigger on phrases like "generate an image", "run this workflow", "ComfyUI", "RunPod", "download model", or "submit workflow".
Best suited for agents with shell/CLI access and file handling (Copilot/Code assistants, CLI-capable agents). The skill expects programmatic execution of the comfy-gen CLI and ability to read/write to network volumes.
ComfyGen is a ComfyUI workflow execution skill for RunPod serverless GPUs. The SKILL.md is comprehensive with detailed workflow analysis, model resolution from HuggingFace/CivitAI, and clear CLI command documentation. No bundled scripts to test. Minor security concern around CLI flag-based secret passing which could leak via shell history, but the skill explicitly warns against displaying API keys. Well-structured and genuinely useful for the ComfyUI-on-RunPod niche.
Documentation-only skill with no scripts. The model resolution workflow is a standout feature — proactively searching HuggingFace for models and presenting findings before downloading is well above average for skill design. Architecture could improve by splitting the long workflow analysis section into references/.