comfy-subgraphs-app-mode

Three frontend features for reusing and simplifying workflows, each version-gated:

Check the running frontend if any of these ever appear missing:

<venv>/bin/python -c "import comfyui_frontend_package as p; print(p.__version__)"

The editor-side mechanics (the JSON shape of definitions.subgraphs[] and its internal link table) live in comfy-workflow-json — that skill is the authority for programmatic subgraph edits. This skill is about the interactive features and what they mean for reuse.

When to Use This Skill

Subgraphs — the basics

A subgraph packages selected nodes into a single collapsible "super node" with exposed input/output slots. Authoring is interactive:

• Create: select nodes → click the subgraph icon in the selection toolbox. ComfyUI auto-exposes the relevant inputs/outputs.
• Edit: double-click empty space inside the subgraph (not on a widget), or use its edit button. A navigation bar shows the current level; Esc or the nav bar exits. Right-click a slot to rename/delete/disconnect; the slot labeled 1 is the spare for adding a new connection.
• Parameters Panel (0.3.66+): select the subgraph and click Edit Subgraph Widgets to reorder/hide its surfaced widgets without entering it.
• Nesting: subgraphs can contain subgraphs; the nav bar tracks depth.
• Unpack: right-click → Unpack subgraph (or the toolbox button) flattens it back to loose nodes.
• Bypass / color / rename: a subgraph behaves like any node.

A plain subgraph lives inside one workflow's JSON (definitions.subgraphs[]). Copy-pasting it into another workflow carries the definition along — reuse-by-copy-paste, fine for one-off porting but not a library.

Subgraph Blueprints — the reusable "library of stages"

This is the answer to "can I reuse a stage across many workflows." Blueprints promote a subgraph into the node library so it can be dragged or searched into any workflow like a built-in node.

• Publish: select the subgraph → click the book (publish) icon in the selection toolbox, or selection-toolbox menu → Add Subgraph to Library. Name it in the dialog (the subgraph's name is the default). It then appears in the node library, searchable and draggable.
• Reuse: drag or search the blueprint into any workflow. Add as many instances as you like.
• Manage: edit/delete blueprints from the dedicated button in the node library (panel section labeled "Comfy Blueprints").

The one caveat that changes how you use them

Instances are isolated snapshots, not live references. Each placed instance is an independent copy you can edit freely; editing one does not affect the others — and crucially, editing the master blueprint does not propagate to instances already dropped into workflows. The docs are explicit that "link instances, synchronous editing" are not yet supported.

Practical consequences:

• Treat a blueprint like a stamp/palette entry, not a symbol with live instances. It's "insert a fresh copy of this stage," not "reference one canonical stage everywhere."
• Versioning is manual. Improve a stage → re-publish (new name, or overwrite — the frontend warns on overwrite). Existing workflows keep their old embedded copy until you delete the instance and drop the new one in. There is no "update all instances."
• Good blueprint candidates are stable, rarely-churning stages (a standard sampler pass, a diffusion core, a reference-latent block, a standard upscale-and-save tail). Volatile experiment stages are poor candidates — you'll be re-stamping constantly.

Where blueprints live on disk

Publishing goes through the userdata API (POST /userdata/<file>/publish), so blueprints are stored server-side (under user/default/ on a self-hosted install) — shared across every browser and workflow hitting that install, not trapped in one browser's local storage. They survive a service restart; they do not live in any individual workflow's JSON.

App Mode — workflow → simple app UI

App Mode (the frontend's internal name is linear mode) hides the node graph behind a purpose-built input/output panel: you pick which node inputs become user controls and which outputs get displayed, and end users just fill the form and hit Run. Available on both self-hosted and Comfy Cloud installs.

• Enter: top-left icon → Enter app mode, or the breadcrumb dropdown at the top of the canvas.
• App Builder (opens automatically the first time, 4 steps):
  1. Select Inputs — click nodes to expose as controls (prompts, sliders, image uploads, model pickers). They preview in the right panel.
  2. Select Outputs — choose which Preview/Save nodes show in the final UI.
  3. Preview — verify the assembled layout.
  4. Set Default View — choose whether the workflow opens in App or Node Graph mode, then Apply.
• Run UI: right-side input panel, a Run button, a left sidebar (rebuild, assets, workflow panel), a Cancel-this-run button during execution, and queue/results toggles. Collapses to a tab view on narrow/mobile screens.
• Save: Ctrl/Cmd + S like any workflow.
• Share: left-sidebar Share → Create a link generates a URL encoding the workflow, layout, and node bindings. Share links are Comfy-Cloud-only — on a self-hosted install you can build and run apps, but a "Create a link" share won't produce a publicly reachable URL.

App config is stored in the workflow JSON

App Mode persists under extra.linearMode and extra.linearData in the workflow JSON (alongside extra.ds, extra.frontendVersion, etc.). This matters for programmatic edits: any transform script that round-trips a workflow must preserve the extra block, or it silently strips the app configuration. If you rebuild a workflow from scratch in a builder script, the app config is gone and has to be reconfigured in the UI. See comfy-workflow-json.

Decision: blueprint vs app mode vs neither