✨ Conduit Apply (Combine & Execute)

Conduit Apply is the node that's supposed to make the whole Conduit optimizer real: you feed it your model, CLIP, and VAE, plug in all the config objects from the other Conduit nodes, and it hands you back an "optimized" triplet to wire into your KSampler. The catch, straight from the source code: right now it's a pass-through. The apply_optimizations method logs which configs it received and returns your model, clip, and VAE completely unchanged. The author's own comment says it best - "For now, pass through… this is a framework for future implementation."

Don't read that as a scam. Read it as honest scaffolding. Conduit Apply is where the pack's plan nodes (Core, Pool, Gate, Path, Seal, Sense) are meant to converge, and it's the first node you'd look at when the pack adds real teeth. Today it earns its keep as the place where the pack's config objects flow - and as a useful way to confirm your plan, pool, and gate nodes are actually connected and being seen.

How it works

Conduit Apply is a fan-in node. Its required inputs are the three things every image workflow has: model (MODEL), clip (CLIP), and vae (VAE). Its optional inputs are the six Conduit config types - conduit_plan, conduit_pool, conduit_gate, conduit_path, conduit_seal, conduit_sense. When the graph runs, it checks which configs are attached and prints a line to the console, something like:

[Conduit] Applied optimizations: Plan: balanced, Precision: auto, Detected: portrait

Then it returns optimized_model, optimized_clip, optimized_vae - which are just the inputs, unchanged. So the output names are optimistic. Wire them into your sampler anyway; the node is designed to sit in that exact spot, and when real optimization lands it'll land behind this interface.

The inputs that matter

There's exactly one beginner decision to make here, and it's not on this node: nothing is configurable on Conduit Apply itself. The only question is which config nodes you attach upstream. Skip the config layer entirely and Conduit Apply is just an expensive bypass - you might as well run the model straight through. The quick-start path from the README is Core → Gate → Apply.

Install

Same as the rest of the pack. ComfyUI Manager, search "comfyui-conduit-optimizer" (or "CONDUIT"), hit install, restart. Or the manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/joe002/comfyui-conduit-optimizer

Then restart ComfyUI. There are no model downloads and no heavy dependencies - the whole pack is pure Python with torch>=2.0 as its only requirement. You should see [CONDUIT] v1.1.0 loaded: 13 optimization nodes in the startup log.

Common issues

The main one is expectations. If you plug Conduit Apply in and your workflow runs at identical speed and produces identical pixels, that's not a bug - it's the current state of the node. The README's "2-3x speedup" framing describes the end state of the framework, not what Apply delivers today. The nodes that actually change your output are the functional ones that sit in the sampling path directly - ConduitPrecisionPatcher, ConduitEmbeddingCache, ConduitLatentCache, and ConduitSpeculativeKSampler. If you want a speedup now, reach for those, and treat Conduit Apply as the dashboard where the plan eventually gets executed.