Load Checkpoint (PIPE)

Stock ComfyUI hands you a Load Checkpoint node with three output dots - MODEL, CLIP, VAE - and from there you're dragging three separate wires across the whole graph. This node collapses that into one. It loads your checkpoint and packs the pieces into a single BASIC_PIPE, so instead of a fistful of wires you run one fat cable to wherever it needs to go. That's the entire pitch, and if your graphs have started looking like spaghetti, it's a good one.

BASIC_PIPE isn't antrobot's invention - it's the five-in-one bundle format that Impact Pack made standard: model, CLIP, VAE, positive conditioning, negative conditioning, all riding one connection. Because this node speaks that same format, its output plugs straight into Impact Pack's own FromBasicPipe / EditBasicPipe nodes and anything else in the ecosystem that eats a basic pipe. It's the same idea as rgthree's Context wire, just aimed at the checkpoint-loading step specifically.

How it works

You pick a checkpoint, it loads it the normal way, then it stuffs MODEL/CLIP/VAE into a pipe and sends it out. The conditioning slots start empty unless you fill them - the node's own description is blunt about it: if positive or negative aren't provided, they'll just be None in the pipe. That's fine, plenty of workflows attach conditioning later. But it means a pipe straight out of this node is a partial pipe until you encode some prompts onto it.

The inputs and outputs that matter

• ckpt_name (required) - the checkpoint dropdown, same list as the stock loader.
• positive / negative (optional) - already-encoded CONDITIONING, not text. If you've got conditioning from somewhere upstream, drop it in here and it travels with the pipe. Leave them empty and they're None.

Two outputs: pipe (the BASIC_PIPE - this is the one you wire onward) and name_string, which is just the checkpoint's filename as text. That second one is genuinely handy - pipe it into your save-filename so your outputs are labeled with the model that made them.

Note the conditioning inputs here take CONDITIONING, not prompt strings. If you'd rather type your prompts right at the loader, that's the sibling node Load Checkpoint With Prompt, which takes text and encodes it for you.

Installing it

No models, no dependencies, no fuss - it's pure Python utility nodes, so this installs in seconds and won't touch your disk budget. Easiest path is ComfyUI Manager → Install Custom Nodes → search "antrobots ComfyUI Nodepack" → Install → restart. Or do it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/antrobot1234/antrobots-comfyUI-nodepack

then restart ComfyUI. The node lands under antrobots-ComfyUI-nodepack/flow-control.

Common issues

The thing that trips people up isn't the install, it's the type. A BASIC_PIPE only connects to nodes that understand BASIC_PIPE. You can't drop this straight into a stock KSampler - you either unpack it first (Impact Pack's FromBasicPipe, or feed it to this pack's own KSampler with Pipe) or you keep everything on the pipe rail. Mixing half-piped, half-loose wiring is where the confusion starts; pick a lane.

The other one: empty conditioning. If your sampler complains about a None positive or negative, that's because you never encoded prompts onto this pipe. Add them at load time, or run the pipe through Encode Conditioning (PIPE) downstream. And since this is a small solo pack - antrobot's personal grab-bag, not a household name with a big issue tracker - if something genuinely breaks, the author explicitly asks you to file a GitHub issue. That's the support channel.