unCLIPCheckpointLoader_motorway_edition

Most checkpoint loaders give you a MODEL, a CLIP, and a VAE. This one adds a fourth output you'll rarely see anywhere else: a CLIP_VISION. That extra piece is what makes the "unCLIP" family of models work - SD 2.1's unCLIP variants and the SDXL-refiner-unclip checkpoints that can take a reference image's embedding and steer generation toward it. This _motorway_edition clone does the same load through the agilly1989 Motorway pipe, storing all four pieces in the shared dict by key.

If you're new to the Motorway: it's a pipe system from the ComfyUI_agilly1989_motorway pack. Start with MotorwayStart, and each _motorway_edition node trades data wires for short key strings - read by key, write by key, all flowing down a single MOTORWAY connection. The clones are auto-generated from ComfyUI's own node registry, so this loader is byte-for-byte the stock unCLIPCheckpointLoader underneath.

What it does. Pick a checkpoint from models/checkpoints that includes a CLIP vision component, and the node loads four things into the Motorway: the diffusion MODEL, the CLIP, the VAE, and the CLIP_VISION output that downstream conditioning needs. This is the loader you'd pair with unCLIPConditioning - the vision output alone does nothing useful until you feed it through that node to inject the image embedding into your conditioning.

The inputs that matter. MOTORWAY is optional here, since a loader has no data inputs - connect an existing Motorway to add to it, or leave it unconnected to start fresh. ckpt_name is the only dropdown. Then there are four key fields telling the node where to store its outputs: OUTPUT_MODEL_key (MODEL), OUTPUT_CLIP_key (CLIP), OUTPUT_VAE_key (VAE), and OUTPUT_CLIP_VISION_key (CLIP_VISION). The single output is the MOTORWAY port. Note the defaults are uppercase - the consumers that read them back default to lowercase (model, clip, vae), so you'll likely be retyping these keys to line them up.

Common issues. Honestly, the biggest issue is that unCLIP checkpoints are rare and old. If you don't have a checkpoint with a built-in CLIP vision, this node is pointless - the dropdown will only contain files that work. Beyond that, the usual Motorway traps apply: keys are exact-match and case-sensitive, and 'MotorwayClass' object has no attribute 'hash_' means you asked for a key that was never stored. And if the node isn't in your menu, the pack's current main branch has the clone registration commented out in __init__.py - re-enable it there or grab a version that had it on.

Install via ComfyUI Manager (search "ComfyUI_agilly1989_motorway") or clone the repo into custom_nodes. No models to download, no dependencies - the pack is pure Python. It's a solo-dev beta ("if things break it's because I broke it"), so treat the Motorway as a tidy-up layer, not the foundation of your workflow.