CreateHookModelAsLoraModelOnly_motorway_edition

Take everything CreateHookModelAsLora does - load a full checkpoint, derive it into a LoRA-style patch, hand you a scheduled HOOKS object - and strip out the text-encoder side. That's CreateHookModelAsLoraModelOnly: a whole checkpoint applied as a model-only LoRA that can be keyframed over the sampling timeline. The _motorway_edition is the core node rehosted on agilly1989's Motorway bus.

When it earns its keep

Model-only patches are the flavor you want when a checkpoint's CLIP weights would fight your prompt - say you're borrowing a model's drawing tendencies (material handling, lighting, anatomy) while keeping your own prompt semantics entirely. Combined with the model-only hook, you get that influence scheduled: strong while the composition sets, faded for the detail pass. Compared to its full sibling, it drops the strength_clip dial and the CLIP processing, which is both simpler and lighter.

Inputs and outputs

• MOTORWAY 🚌💨 - the bus, required.
• INPUT_prev_hooks_key - key of an existing HOOKS group to append to, default prev_hooks.
• ckpt_name - the checkpoint dropdown (files in ComfyUI/models/checkpoints), set directly on the node.
• strength_model - model strength, -20 to 20, default 1.
• OUTPUT_HOOKS_key - key for the HOOKS object, default HOOKS.

Only output: the Motorway.

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/agilly1989/ComfyUI_agilly1989_motorway

Restart, or ComfyUI Manager → "ComfyUI_agilly1989_motorway". No pip deps; you supply the checkpoint files.

Caveats

Pack-wide: active beta, "BIG BROKEN WITH ASYNC" README banner, _motorway_edition clones commented out of __init__.py in the checked version (re-enable ClonedNodeMapping if missing), keys exact and case-sensitive. Same caveats as its siblings, condensed: hooks are EXPERIMENTAL in core ComfyUI, so confirm your sampler honors HOOKS or this degrades to a static model-only patch; and it's heavier than a real LoRA load because the whole checkpoint is read to derive the patch. If you don't need scheduling or the full-model blend, a plain LoraLoaderModelOnly is lighter and simpler.