UNET CLIP VAE To Model Info

Sage Utils has a rule: selectors pick and describe, loaders load. UNET CLIP VAE To Model Info is where that rule turns into something useful - it bundles the three selector outputs into a single model_info value so your workflow can treat a set of separately-chosen parts like one virtual checkpoint.

The inputs are exactly the three MODEL_INFO outputs you've already built with the selector family: unet_info from UNET Selector, clip_info from a CLIP selector, vae_info from VAE Selector. The node does almost nothing to them - its execute method literally bundles the three into one tuple and hands it back as model_info. That's not laziness, that's the architecture. The work is in the type: everything downstream that accepts a MODEL_INFO now sees a complete, checkpoint-shaped description, without ever having loaded a single weight.

So when do you actually want this? When your model is genuinely assembled from parts. A distilled UNET plus a stock VAE. A Flux UNET with a custom CLIP and a swapped VAE. Anything where you've deliberately kept the components separate instead of using a monolithic checkpoint file - which is increasingly common now that people run fp8 UNETs and quantized VAE variants. The model_info output can flow into Sage Utils' metadata nodes (Construct Metadata and friends), which is the pack's whole reason to exist: your final saved image should record which UNET, CLIP, and VAE produced it, and this node is what makes that recording natural.

One honest caveat: on its own, this node doesn't load anything. You've bundled descriptions, not weights. To actually run a generation you'll still route the parts to the pack's loaders - Load UNET, Load CLIP, Load VAE - or to Load Model From Info if you want the whole bundle loaded in one step. Think of it as the "packaging" step in a build-your-own-checkpoint pipeline, not the runtime step.

And a naming note, since it'll trip you up once: the node's description says "Returns a list with the unets, clips, and vae in it to be loaded," and its model_info output is type MODEL_INFO, which is the same type the selectors emit. If you find yourself feeding this node's output back into a loader that expects a single component, check which info type that loader actually wants - the whole family shares the MODEL_INFO name but the internal slots differ.

Install is the pack standard - ComfyUI Manager (search "Sage Utils") or:

cd ComfyUI/custom_nodes
git clone https://github.com/arcum42/ComfyUI_SageUtils.git

then restart. No model downloads, no heavy dependencies. It's pure glue, and that's exactly what it should be.