Controlled ModelPatch Loader

Model patches are ComfyUI's way of tweaking a diffusion model without a full new checkpoint - a small patch file that layers onto the model at load time. The stock ModelPatchLoader (from comfy_extras) handles them; this node wraps it with a trigger input and marks itself EXPERIMENTAL, which tells you most of what you need to know about where it sits in the pack.

Fire the trigger and it loads the patch and outputs a MODEL_PATCH; leave it empty and you get None instead. The MODEL_PATCH output feeds into the model_patch input that the newer model-loading nodes accept, applying the patch to your model. That's the whole feature set - the source is the usual thin delegation to ModelPatchLoader.load_model_patch() with a gate in front.

The honest assessment: this is the least likely node in the pack to change your life. Model patches are a niche feature most workflows never touch, and the pool of MODEL_PATCH-consuming nodes is small compared to, say, every node that eats a MODEL. The trigger buys you the same thing it buys everywhere here - skip the load when the branch isn't running, keep a heavy patch out of memory when you don't need it - but you're adding experimental wrapper behavior to an already-niche feature. If you don't already know why you'd need a gated model patch, you don't need this node.

The inputs that matter

Inputs are the name picker for the patch file and the trigger, with the usual warning: a paused node emits None, and anything downstream expecting a real MODEL_PATCH will error out rather than gracefully skip.

Install

Install is shared with the pack - ComfyUI Manager (search "DeleteModelPassthrough") or clone into custom_nodes and restart. No downloads beyond the pack itself, which carries only psutil on top of torch. If you hit a snag, remember the pack's own README calls everything beyond the delete nodes experimental - treat console output as the documentation.