Smart Model Info

Smart Model Info answers a boring but constant question: what model is this, actually? You feed it a MODEL object and it prints the architecture class, the UNet config, the latent format, and any model options or attachments - as a readable string in the UI - then passes the model through untouched.

It's a curiosity and debugging node, the kind of thing you reach for when a model loads but behaves oddly, or when you grabbed a workflow off the community and want to confirm the checkpoint really is what the title claims. ComfyUI models are opaque objects; this node is a window into them.

How it works

It inspects the ComfyUI model wrapper rather than the raw file. For each section it can read, it appends to a text block:

• Model config - the config class name (this is the closest thing to "architecture" the wrapper exposes), the UNet config dict (channels, attention settings, etc.), and the latent format class.
• Model options - the wrapper's model_options, which is where things like custom attention or patching options live. Useful for spotting when a model has been modified by upstream nodes.
• Attachments - any custom data attached to the model, which some node packs use to stash metadata.

Whatever it can't read, it skips gracefully or labels unknown. The model output is a straight passthrough, so you can drop the node into an existing wire and read the info without disturbing the pipeline.

The inputs and outputs that matter

• model - the MODEL to inspect.

Outputs: model (passthrough, wire it onward) and info (STRING, the report). It's an output node, so the report also renders in the UI.

What it's actually for

Two realistic uses. Debugging: a diffusion model that produces garbage often has a tell in its config - wrong latent format, unexpected attention config - and a glance at this output beats guessing. And workflow vetting: when you load a downloaded workflow and it pulls in a model you don't recognize, this tells you the architecture class before you burn a render on it. It's the kind of node you use five times, then keep in a corner of a utility workflow for the next time.

Installing it

It's one of ~30 nodes in SmartHelperNodes; install the pack once:

cd ComfyUI/custom_nodes
git clone https://github.com/slvslvslv/ComfyUI-SmartHelperNodes

Restart ComfyUI, or use ComfyUI Manager ("SmartHelperNodes"). No pip dependencies, no model downloads.

Common issues

Set expectations: the report is only as deep as the model wrapper's config exposes. For safetensors-loaded diffusion models the config can be generic (the wrapper doesn't always carry the source filename), so don't expect it to always name the exact checkpoint - it tells you the architecture and structure, not necessarily the file that produced it. If you need the filename, the pack's SmartModelLoader outputs model_name from the load instead. And because it's passthrough, an unconnected model input just yields a "no model information" string - harmless, but wire it before you trust it.