Pipe to Checkpoint Models

This node is the exact inverse of AV_CheckpointModelsToParametersPipe. That one takes a checkpoint, a VAE, a couple of upscalers, and up to three LoRAs and packs their names into one PIPE. This one takes that PIPE and gives you every one of those names back, as separate outputs, so the rest of your graph can wire them into actual loaders.

It's the same manifest-and-unpack pattern the whole ArtVenture pipe system runs on: build the bundle once, carry it as a single wire through the parts of your workflow that don't care what's in it, then break it back apart exactly where you need the individual pieces.

How it works

Feed it a PIPE and it reads out the model-selection fields that were packed into it, unchanged. It doesn't load anything and doesn't check that the named files exist - it's purely handing back strings (and one nested PIPE) for something downstream to act on.

The inputs and outputs that matter

One required input:

• pipe - the bundle to unpack, typically built upstream by AV_CheckpointModelsToParametersPipe.

Nine outputs, all pass-through reads of the pipe's contents:

• pipe - the same bundle again, so you can keep unpacking other parts of it elsewhere (prompts via AV_ParametersPipeToPrompts) off the same wire.
• ckpt_name / secondary_ckpt_name - the primary and (if set) secondary checkpoint names.
• vae_name - the VAE name.
• upscaler_name / secondary_upscaler_name - the two upscaler names.
• lora_1_name, lora_2_name, lora_3_name - up to three LoRA names.

Every one of these is a name, not a loaded model - you wire them into the pack's own loader nodes (or CheckpointLoader/VAELoader/LoRA loader equivalents) to actually get a MODEL, VAE, or LoRA object.

How to install it

Part of the Art Venture pack. ComfyUI Manager: search comfyui-art-venture, install, restart. Manually:

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

Restart ComfyUI. No model downloads - the models it names still need to exist in your usual folders; this node just hands their names onward.

Common issues & troubleshooting

Every output is None or empty. The pipe you fed in was never built by something that set these fields. If it came from AV_PromptsToParametersPipe alone (the prompts-only builder) with no model selections chained in, there's nothing here to unpack - go back to AV_CheckpointModelsToParametersPipe and confirm its output actually reaches this node's pipe input.

A name comes out but the loader downstream errors. This node doesn't verify anything - it hands back whatever string was packed in, even if that checkpoint or LoRA no longer exists on disk. Check the actual file is present in the matching models/ subfolder; the failure always shows up one step later, at the loader.

Only using one or two of the nine outputs. That's normal and fine - leave the rest unwired. You don't have to consume every field just because the node exposes it.

This feels like more machinery than a normal workflow needs. For a single generation, a plain Load Checkpoint node is simpler. This pays off specifically in large templated or automation-driven graphs, where an external caller swaps the whole model selection by feeding a different pipe into one place instead of editing several loader nodes scattered through the graph.