REPO PACKAGE
runcomfy-com skills
COMMUNITY30skills
0ready to use
LABSCO SUMMARY
Sixteen of the thirty route across several backend models depending on what the request actually needs — ai-image-generation, ai-video-generation, ai-music, ai-avatar-video, image-edit, image-to-video, video-edit, face-swap, lipsync, relight, controlnet-pose, image-inpainting, image-outpainting, video-extend, video-inpainting, and video-outpainting each pick among named models such as Kling, Wan, Seedance, Nano Banana, GPT Image 2, Flux, and HappyHorse. Twelve more call one named model directly and ship that model's own documented prompting patterns — flux-2-klein, flux-kontext, gpt-image-2, gpt-image-edit, happyhorse-1-0, kling-3-0, nano-banana-2, nano-banana-edit, seedance-v2, wan-2-7, ace-step, and elevenlabs-music-generation. The remaining two are runcomfy-cli, which teaches the install-auth-invoke workflow every other skill depends on, and codex-pet, an outlier that builds a desktop-pet spritesheet for OpenAI Codex out of a single GPT Image 2 call.
Separately, the README describes a Cursor and Claude plugin at plugins/runcomfy that bundles RunComfy's hosted MCP endpoint into four focused skills — image generation, video generation, ComfyUI workflows, and LoRA training — installable through the shared Claude Code/Cowork plugin format or the Cursor marketplace. This is for someone who already pays for RunComfy and wants an agent that picks the right image, video, or audio model for a request rather than a way to generate media for free; the repository itself is MIT-licensed, so reading or forking the skill and plugin definitions costs nothing even though running any of them does.
READ THE FULL ANALYSIS
The README admits its own skill docs can go stale. It closes with a direct warning that "some older skill documents retain historical installation links and model defaults" and tells a reader to verify current model inputs and prices before running jobs — RunComfy's own acknowledgment that some of the thirty files described above may still point at superseded commands or old pricing, which we did not check file by file.
The plugin's four skills are not part of the thirty counted above. Our skill index only tracks the top-level directories under the README's "Individual CLI skills" section; the plugin's own README, install steps, and example prompts live at plugins/runcomfy and were not in the material we reviewed, so we cannot describe how those four actually behave, only that the top-level README says they exist.
ALSO IN THIS PACKAGE
PLUGINS
1 itemPLUGIN
runcomfy
Run ComfyUI workflows in the cloud on GPU, generate images with FLUX and Seedream, create videos with Seedance, Wan and LTX, and manage LoRA training.
MCP SERVERS
1 itemMCP SERVER
runcomfy
http · https://mcp.runcomfy.com/mcpWHAT'S INSIDE
30 showing · 30 total30 of 30
SETUP
Showing all 30 skills
ace-step
Makes music from a short written brief — genre, mood, instruments, and lyrics if you want singing — at a fraction of a cent per second.
ai-avatar-video
You bring a face and a voice, and it works out which of RunComfy's models will make the mouth and the gestures match, then hands back the finished video.
ai-image-generation
Makes and changes pictures from a written description. RunComfy carries dozens of image models that are each good at different things, and this reads what you asked for and sends it to the one that fits.
ai-music
Music on demand — it weighs what you asked for against what you want to spend, and sends the job either to the studio-grade model or to the one that costs a tiny fraction of it.
ai-video-generation
Video from a sentence or from a single photo, with the choice of model — fast and cheap, photoreal, or cinematic — made for you out of what you asked for.
codex-pet
Builds a custom pixel-art pet that animates on your desktop while OpenAI Codex works, generating the sprite from one reference photo with RunComfy's image model instead of Codex Pro's built-in pet generator.
controlnet-pose
Takes the movement out of one video and puts it onto a different character — the dance stays the same, the dancer changes.
elevenlabs-music-generation
The polished end of AI music: a mixed, release-ready track built from a written brief, up to five minutes long, with or without singing.
face-swap
Puts one person's face onto the person already in a photo or a video, with the model behind it chosen according to whether it is a still, a batch, or moving footage.
flux-2-klein
An image generator built for speed: a picture comes back in about a second, so you can try something, look at it and try again, at the cost of a little final polish.
flux-kontext
The precise one for editing a single photo: no masks, no layers, just one short sentence saying what should be different.
gpt-image-2
An image model that gets the lettering right — headlines, labels, packaging, signage in other languages — and follows a detailed instruction closely.
gpt-image-edit
Changes an existing photo on instruction, and is the one to pick when the photo has writing in it, or a layout that has to survive the edit.
happyhorse-1-0
Type out a scene and get a 1080p clip back with its own synchronised sound — currently the top-rated video model in blind side-by-side voting.
image-edit
Photo editing without having to know which tool does what: say what should change, and the job goes to whichever of RunComfy's edit models is strongest at that kind of change.
image-inpainting
Paint over the part of a photo you want gone or replaced, say what should be there instead, and only that patch is redone.
image-outpainting
Extends a photo's canvas beyond its original edges on RunComfy — changing aspect ratio or revealing more scene — while keeping the existing content untouched.
image-to-video
A still photo goes in, a moving clip comes out; which of the video models does the work depends on what you want it to do.
kling-3-0
Cinematic video from a written scene or from a photo, at three price-and-quality levels, with the sound made alongside the picture rather than added afterwards.
lipsync
Whatever you start from — footage you already shot, a single portrait, or nothing but a script — the mouth ends up moving in time with the words.
nano-banana-2
Fast, cheap images for the early end of a job: drafts to choose between, variants, thumbnails — not the final hero shot.
nano-banana-edit
Changes an existing photo while keeping whoever, or whatever, is in it looking like the same person or the same product.
relight
Changes the lighting of a still photo on RunComfy — direction, color temperature, mood — without redoing the shot or touching anything else in the frame.
runcomfy-cli
The command-line tool every other RunComfy skill runs on: it signs you in, waits while the job runs, and saves the finished file to your machine.
seedance-v2
Makes one short cinematic shot out of a description plus whatever you have to show it — photos of the character, clips for the look, a voice to copy.
video-edit
Looks at what kind of change you want to make to a video you already have — a general restyle, motion copied from a reference clip, or a lightweight outfit swap — and picks the model built for that job.
video-extend
Takes a video clip you already have and generates more footage that continues it, using Google's Veo 3-1 model to keep the motion, lighting, and characters consistent with the original.
video-inpainting
Removes or replaces a specific area of a video — a watermark, a stray wire, a person passing through the background — using a written description of the spot instead of a hand-drawn mask.
video-outpainting
Expands the visible area around a video's edges — for example turning a vertical clip into a widescreen one — while keeping the original action centered and consistent.
wan-2-7
Turns a text prompt into a video using Wan-AI's Wan 2.7 model, with the option to sync the character's lips to an audio track you already have.
No skill in this package matches that.
HOW TO GET IT
WHOLE PACKAGEInstall the whole package (30 skills)
npx skills add runcomfy-com/skillsSINGLE SKILLOr install a single skill
npx skills add runcomfy-com/skills --skill <name> --full-depthPick the skill name from the Skills tab — each entry there installs independently.
PLUGIN MARKETPLACEAdd the marketplace, then install the plugin
/plugin marketplace add runcomfy-com/skills/plugin install runcomfy@runcomfyTyped inside the agent's own prompt, not in a terminal. The marketplace is called runcomfy, which is the part after the @.