Workflow Guide Generator

You've downloaded some 200-node workflow, stared at the spaghetti for ten minutes, and still can't tell if it's a Flux upscale pipeline or a video-to-EXR VFX pass. The Workflow Guide Generator node exists for exactly that moment. You drop it on the canvas, click Generate Artist Guide, and a big text panel appears off to the side explaining what the workflow does, what models it uses, what sampler and step counts it's set to, and roughly what it's for. It's a documentation-on-demand tool, not a generation node.

The name is slightly misleading in the best way: it doesn't call any API, needs no key, and uploads nothing. All analysis happens locally inside ComfyUI.

How it actually works

The Python side of this node is a stub - no inputs, no outputs, OUTPUT_NODE = True, and an execute that returns nothing. Don't panic when you see zero sockets; that's by design. The real machinery is split between the frontend extension and a local HTTP endpoint the pack registers on ComfyUI's own server.

When you click the button, the JavaScript serializes your entire graph (app.graph.serialize()), POSTs that JSON to /workflow-guide/generate on localhost, and a backend WorkflowDocumentationEngine runs nine sub-engines over it: a workflow analyzer, a purpose engine, a flow engine, a diagram engine, model discovery, parameter extraction, and a node-purpose explainer. The response comes back as a big JSON document, the frontend deletes any existing guide panel, and it spawns a fresh Workflow Documentation Panel next to your graph with everything stuffed into it.

How it "understands" your workflow is worth knowing before you trust it. It's keyword and node-type heuristics, not magic. Model discovery walks every node and matches loader names - CheckpointLoaderSimple, UNETLoader, CLIPLoader, VAELoader, LoraLoader, ControlNet loaders, upscale nodes - and grabs the first widget value as the filename. The purpose engine joins all node type names into one string and checks for patterns like vhs_loadvideo + saveexrframes (video to EXR) or video anywhere (video processing). The sampler, scheduler, steps and CFG get pulled from the widgets of the KSampler-family nodes it recognizes. Anything it can't map becomes "Not Detected" or a generic description.

What you actually do with it

Add the Workflow Guide Generator node anywhere on the canvas (it doesn't need to be wired to anything - you can't wire it to anything), click the button, wait a second, and read the panel it creates. That's the whole workflow. It's read-only: nothing in your graph is modified, which the pack's FAQ is careful to promise.

The output panel - rendered by the sibling WorkflowDocumentationPanelNode - shows purpose, an overview, main model / text encoder / VAE, LoRAs, ControlNets, sampler, scheduler, steps, CFG, resolution, a stage-by-stage flow like Load Models → Process Prompt → Generate Image → Save Result, and a list of key nodes with plain-English explanations. You can reposition the panel next to the workflow you're documenting, save it as part of the workflow JSON, and share it along with the graph - which fits nicely with the ComfyUI norm of embedding the workflow in the output PNG.

Installing it

Grab it via ComfyUI Manager (search "Workflow Guide Generator") or clone it directly:

cd ComfyUI/custom_nodes
git clone https://github.com/GowthamGENAIVFX/ComfyUI-Workflow-Guide-Generator

Restart ComfyUI, then look under the Workflow Documentation category. There are no extra Python packages to install - requirements.txt is explicitly empty ("No external dependencies required"), and it leans on aiohttp, which ships with ComfyUI's server. No model files to download either. It does want Python 3.10+ and a recent frontend.

Where people get burned

The heuristics are the ceiling. If your workflow uses custom loader nodes the pack doesn't recognize, expect a lot of "Not Detected" fields and generic node blurbs. The panel spawns to the left of your graph's leftmost node (about 1150px over), so on a wide workflow it can land partly off-screen - just drag it back. If the panel doesn't appear after installing, the pack's own docs say hard-refresh with CTRL+F5 to clear the frontend cache.

Also be honest about what this is right now: it's an early, single-commit project, and there's essentially zero community chatter about it - treat the guide as a helpful starting point for onboarding into an unfamiliar workflow, not as a substitute for checking the actual node settings yourself. For that job it's genuinely handy, and it's free.