InstructPixToPixConditioning_motorway_edition

InstructPixToPixConditioning_motorway_edition wraps core's InstructPixToPixConditioning, the node that prepares conditioning for an InstructPix2Pix model - the "edit this image according to this instruction" model family. The wrapper pulls its inputs from the Motorway by key and writes its outputs back to keys, same as every _motorway_edition node in this pack.

What it's actually for

InstructPix2Pix models are trained to take an input image plus a text instruction ("make it night," "turn the dog into a cat") and produce an edited version. The conditioning node is the translation layer: it encodes the input image, packs it with your positive/negative text conditioning, and produces a latent the sampler can work from. There's a hard prerequisite that trips people up: you need a checkpoint that was actually trained for InstructPix2Pix - ip2p models (e.g. the original instruct-pix2pix SD 1.5 checkpoint) - not a general-purpose model. Feed a normal SDXL checkpoint into this node and you'll get nonsense, because the model never learned the instruction-conditioned task. The KB's 2026 sampler landscape doesn't have much warm to say about the ip2p family - it's an older, niche approach that mostly lives on in specific legacy workflows - so before you build a pipe around it, confirm the model you're loading actually is an ip2p one.

The motorway wiring

All five inputs are key fields, no widgets at all:

• INPUT_positive_key / INPUT_negative_key (defaults "positive" / "negative")
• INPUT_vae_key (default "vae")
• INPUT_pixels_key (default "pixels") - the image to be edited
• Outputs: OUTPUT_positive_key / OUTPUT_negative_key (defaults "positive" / "negative"), OUTPUT_latent_key (default "latent").

The outputs overwrite the same keys they read from, so a downstream KSampler_motorway_edition reading "positive"/"negative" gets the conditioned versions, and the encoded latent is waiting under "latent". Only output socket is MOTORWAY 🚌💨; pull keys back out with Motorway AxB ramps when something needs a real cable.

Install and caveats

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

Restart ComfyUI, or install via ComfyUI Manager. No pip deps; the pack downloads no models - the ip2p checkpoint is on you.

The standing warnings apply. This pack is a one-developer beta: the README says things break because the author broke them, and the current repo HEAD has the _motorway_edition registration commented out (a hotfix mid-refactor), so a fresh clone may not even show the node. The 'MotorwayClass' object has no attribute 'hash_' runtime error means a key you referenced was never written to the pipe. And the practical caveat that predates all of it: if your workflow isn't already running an actual InstructPix2Pix model, this node has nothing to do. The stock node is identical without the pipe, and for a niche niche like this, that's usually the version you want.