Builds reproducible ComfyUI environments into versioned releases on the Comfy developer platform.
npx skills add https://github.com/Comfy-Org/comfy-cli --skill comfy-build
Please help me install this Agent Skill. Skill: comfy-build Source: https://github.com/Comfy-Org/comfy-cli/tree/main/comfy_cli/skills/comfy-build Command: npx skills add https://github.com/Comfy-Org/comfy-cli --skill comfy-build
Packaging a working ComfyUI setup — its custom nodes, models, Python dependencies, and ComfyUI version — into a reproducible, deployable environment is error-prone and manual. This Skill turns a local install, a Desktop snapshot, a workflow JSON, or a written description into a committed comfy-build.yaml spec and a green, deployable release on the Comfy developer platform.
Ask the assistant to build a reproducible ComfyUI environment from your local install using comfy build init, then validate, push, and cut a release for the linux/nvidia target.
High-intent search queries and answers about installing and using this skill.
Run comfy build init on your install directory to scan it and write a comfy-build.yaml spec, then validate, push, and cut a release with comfy build release create. Pass --python pointing at the install's venv for split or Desktop layouts.
Use comfy build init with --from-workflow pointing at the workflow file, which writes a local spec without creating a Build. Models named in the workflow are not included, so resolve each with comfy build refs resolve and set a ComfyUI version before cutting.
The most common cause is init copying the entire pip freeze into pipDependencies, which the builder applies as overrides that can conflict with the base image's torch. Cut the first build with pipDependencies emptied, keeping only deliberate pins you can justify.
Yes, use comfy build init with --from-snapshot pointing at the Desktop snapshot JSON. The import requires cloud sign-in and carries no models, so use a direct scan instead when private model weights must be included.
A release is deployable when its linux/nvidia artifact reaches ready with an image ref, regardless of the rolled-up status. A release can show complete with failed targets, so check the deployable flag and per-target artifact states rather than status alone.
Do not use it to deploy environments — it stops at a green release and hands off to comfy-deploy, which involves ongoing spend. It also requires the platform beta to be enabled on your account, and sign-in for any builder-facing operation.
jackspaceGuide ComfyUI workflow creation, node configuration, and optimization.
nadicodeaiDeploy high-throughput LLM inference on NVIDIA GPUs with tensorrt-llm.
imgamerDetect cross-document inconsistencies between GDDs and the entity registry.
NousResearchGenerate images, video, and audio via ComfyUI diffusion workflows.
