Deploy ComfyUI build releases as serverless endpoints using comfy-cli deploy commands.
npx skills add https://github.com/Comfy-Org/comfy-cli --skill comfy-deploy
Please help me install this Agent Skill. Skill: comfy-deploy Source: https://github.com/Comfy-Org/comfy-cli/tree/main/comfy_cli/skills/comfy-deploy Command: npx skills add https://github.com/Comfy-Org/comfy-cli --skill comfy-deploy
Running a ComfyUI build as a hosted, serverless endpoint involves cost traps and state complexity: deployments bill continuously while holding compute, new releases create duplicate billing deployments, and failed stops can silently keep charging. This Skill guides an AI agent through the comfy deploy command group so deployments are created, scaled, monitored, and torn down without surprise bills.
up, submit API-format workflows with run, and pause or tear down compute with stop, start, scale, and delete.queued, provisioning, starting, ready, unhealthy), the supersedes warning on duplicate deployments, and the --min standing charge so users consent before spending.status, events, and logs to identify why a deployment failed, and maps deploy_* error codes (like deploy_job_submit_unknown or deploy_payment_required) to the correct recovery move.comfy deploy refs compute for available GPU/region pairs, discloses the continuous billing model, runs comfy deploy up --watch, submits a workflow with comfy deploy run, and stops the deployment when finished.Ask the agent to deploy the current Comfy Build release as a serverless endpoint and run your API-format workflow against it, then stop the deployment when done.
High-intent search queries and answers about installing and using this skill.
Run comfy deploy up with a GPU class and region taken from comfy deploy refs compute, after confirming a deployable release exists with comfy build release show. The deployment must reach the ready state before comfy deploy run will submit workflows to it.
Use comfy deploy run with an API-format workflow JSON exported via ComfyUI's File → Export (API). The deployment must be in the ready state, and local files are uploaded only from the install's models/, input/, and output/ directories or paths passed via --asset-root.
A stop_failed status means the stop did not take and compute may still be billing; run comfy deploy stop again with the deployment id. Also check comfy deploy ls for older deployments of the same Build, since a new release creates a second deployment rather than replacing the first.
Stop pauses the deployment while retaining its endpoint URL and staged models, so start resumes the same deployment. Delete enqueues teardown and soft-deletes the record irreversibly; serving again requires up to create a new deployment with a new URL.
No, GPU and region are immutable on a live deployment and changing them returns deploy_immutable_compute. Stop the deployment first, then use comfy deploy scale --gpu or --region, and start it again.
It means the submission timed out and the job may or may not have been created, with no API to confirm either way. Do not resubmit, since each run uses a fresh idempotency key and a retry would be a second billed job; check comfy deploy status and report the ambiguity to the user.
XyHalcyonAutomate image, video, and audio generation with ComfyUI workflows.
Liberty91LTDSubmit URLs to the URLScan.io API and retrieve scan results.
oci-ai-incubationsMonitor Kubernetes cluster health and readiness across all namespaces every 30 seconds.
jleechanorgAutomate ComfyUI generative workflows with parameter injection and local or cloud routing.