Checkpoint Models to Pipe

This node gathers a pile of model choices - a checkpoint, a VAE, upscalers, up to three LoRAs - and packs them into a single PIPE bundle that travels through your graph as one wire. It's part of ArtVenture's "parameters" system, the pack's answer to the spaghetti problem: instead of running a dozen separate wires for every model your workflow uses, you collect the selections here and carry them as one tidy object.

The important thing to get straight: this bundles names, not loaded models. It's a manifest - "use this checkpoint, that VAE, these LoRAs" - assembled in one place and passed downstream, where the rest of the ArtVenture parameter system reads the bundle and actually loads things. That indirection is the point. It's built for big, templated, reused workflows where you want all the model configuration collected at one node instead of scattered across the graph, and where an external caller can swap the whole set by feeding different selections.

How it works

You pick a primary checkpoint and, optionally, a stack of related pieces - a second checkpoint, a VAE, two upscalers, three LoRAs. The node writes all of that into a PIPE object. It can also take an existing PIPE in and extend it, so you can build the bundle up across nodes rather than filling every field on one. Downstream, the parameter system unpacks the PIPE to drive loading.

Inputs and outputs

One required input:

• ckpt_name - the primary checkpoint. The dropdown lists whatever's in your checkpoints folder.

The optional inputs are all bundle extras, each defaulting to None:

• pipe - an existing PIPE to extend rather than start fresh.
• secondary_ckpt_name - a second checkpoint (for refiner-style or two-model setups).
• vae_name - a VAE to include.
• upscaler_name / secondary_upscaler_name - upscale models to carry along.
• lora_1_name, lora_2_name, lora_3_name - up to three LoRAs.

The single output is a PIPE - into the downstream ArtVenture parameter nodes that consume it.

Installing it

Comes with the ArtVenture pack. ComfyUI Manager: search comfyui-art-venture, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/sipherxyz/comfyui-art-venture

Restart ComfyUI. No downloads - but the models you name in the bundle need to actually exist in their folders, since something downstream will try to load them.

Common issues

"It didn't load my checkpoint." Right - this node doesn't load anything. It builds a bundle of names. If nothing downstream unpacks the PIPE, nothing happens. This node only makes sense as part of the ArtVenture parameter chain; on its own it's inert.

Dropdowns are empty or short. The optional selectors read from your model folders. If lora_1_name only shows None, you have no LoRAs installed where the pack looks. Same for VAE and upscalers.

It's more machinery than you need. For a normal workflow, a plain Load Checkpoint node is simpler and clearer. The PIPE bundle earns its keep in large, templated, automation-driven graphs where consolidating model config into one wire actually pays off. If you're not building that, you probably don't want this node.

Named a model that isn't there. The bundle stores the name regardless; the failure shows up downstream when the loader can't find the file. Confirm every named model exists before wondering why the load step errors.