Google AI - Workflow Analyzer

You download a workflow from the internet, load it in ComfyUI, and it's a wall of nodes you've never seen. Or a friend sends you a .json and asks "why is this broken?" This diagnostic node exists for exactly that moment: hand it the workflow JSON and it gets Gemini to explain what the graph is doing - what each node is, what's being loaded, how the pieces connect.

The mechanism is a neat two-step. First it parses the JSON you give it - and this is the clever bit, it handles both formats ComfyUI workflows come in: the raw API format (a dict of node IDs with class_type fields) and the UI format (nodes[] with type fields), plus it accepts either pasted JSON or a filesystem path to a .json file. It collects every unique node type into a list, then sends that list to Gemini with a prompt asking for analysis. What comes back is the analysis_report string.

The inputs that matter

• workflow_json - the only required input. Paste the raw workflow JSON, or a path like C:\downloads\myworkflow.json.
• model - defaults to gemini-3.1-pro-preview. If you just want a fast read, the flash models in the combo are cheaper and plenty good at this task.
• api_key - leave empty; resolved from .env as GEMINI_API_KEY.

Output: one analysis_report STRING. It's a text node, so wire it to a text display widget and read it there.

Where this is (and isn't) useful

Honestly? It's a debugging and learning aid, not a build tool. It's genuinely handy when a workflow won't load because a node type is missing - the report tells you what's referenced, so you know what pack to install. It's also a decent first pass at understanding someone's complex graph before you start deleting nodes. What it won't do is fix the workflow or check your wiring; it's reading a node list, not executing anything. If you feed it a malformed JSON it just returns an error string rather than crashing, which is the right behavior for a diagnostic node.

Installing it

ComfyUI Manager → search "COMFYUI_PROMPTMODELS" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/cdanielp/COMFYUI_PROMPTMODELS

Add GEMINI_API_KEY=AI... to the pack's .env and restart. ComfyUI >= 0.26.0 required; dependencies are requests and Pillow.

Gotchas

Key hygiene is the recurring theme of this pack: .env, not the node field, or the key ends up in your workflow JSON and PNG metadata. And remember the output quality is only as good as the model - with the default Pro model and the full node list of a big workflow, the response can get long and rambly. Ask it targeted questions by trimming the JSON first if you want a focused answer. Also, this one's a paid Gemini call per run - free tier gives you a little, but repeated debugging sessions will burn credits.