unCLIPConditioning_motorway_edition

Normal text conditioning tells the model what you want in words. unCLIPConditioning is the alternative path: it takes an image's CLIP vision embedding and injects that into the conditioning, so the model gets "roughly like this" as an instruction. It's the mechanism behind the old SD 2.1 unCLIP and SDXL-refiner-unclip style workflows, and this _motorway_edition clone runs it through the agilly1989 Motorway pipe - which is actually a natural fit, since it needs two things (conditioning + a CLIP vision output) that you'd otherwise be dragging wires around for.

Quick Motorway primer, in case you hit this page cold. It's a pipe system from the ComfyUI_agilly1989_motorway pack: start with MotorwayStart, then every _motorway_edition node reads its inputs from and writes its outputs to a shared dict via short key strings, all threaded down one MOTORWAY connection. These nodes are auto-generated clones of core ComfyUI nodes, so the math is identical to stock.

What it does. It takes your CONDITIONING and a CLIP_VISION_OUTPUT and appends the image embedding to it with two knobs. strength (default 1.0, range −10 to 10) is how hard the image pushes against the text prompt; set it to 0 and the node is a no-op passthrough. noise_augmentation (default 0.0, range 0 to 1) adds noise to the CLIP vision embedding before injection - more noise means the output drifts further from the source image while still borrowing its general composition. Zero means "follow the reference closely."

The inputs that matter. MOTORWAY is required, so this node has to sit in an existing Motorway. INPUT_conditioning_key (default conditioning) is where your prompt-encoded conditioning lives in the dict, and INPUT_clip_vision_output_key (default clip_vision_output) is where the CLIP vision output lives - remember the unCLIP checkpoint loader stores it under CLIP_VISION uppercase, so you'll likely retype this one. OUTPUT_CONDITIONING_key (default CONDITIONING) is where the enriched conditioning is stored for the sampler. Output: just the MOTORWAY port.

Common issues. The key/case mismatch is the #1 failure here: clip_vision_output lowercase by default, CLIP_VISION uppercase if you let the loader use its defaults. Wrong key, and you get the pack's signature error, 'MotorwayClass' object has no attribute 'hash_' - that's "key not in the Motorway," not a crash you should fear. If you're pushing strength past ~1.5 or turning up noise_augmentation, results get unpredictable fast; this node is from an experimental era of the ecosystem and behaves like it. And if the node is missing from your menu entirely, the pack's current main branch has the clone registration commented out in __init__.py.

Install: ComfyUI Manager (search "ComfyUI_agilly1989_motorway") or clone into custom_nodes, restart, no models to fetch. Solo-dev beta pack - "it works on my system" is the author's own standard, and for a niche conditioning trick like this, that's about the right bar.