Load Checkpoint With Prompt (PIPE)

This is the "load the model and set up my prompts in one node" shortcut. In stock ComfyUI, getting to a ready-to-sample state means a Load Checkpoint node plus two CLIP Text Encode nodes plus the wires between them. This folds all of that into a single box: pick a checkpoint, type your positive and negative prompts right there, and out comes a fully-loaded BASIC_PIPE with the conditioning already encoded. It's the fastest way to go from nothing to a sampleable bundle.

BASIC_PIPE is the model-CLIP-VAE-positive-negative bundle that Impact Pack popularized - one wire instead of five - so what this node emits interoperates with the rest of the pipe ecosystem. Think of it as the front door to a piped workflow: load here, and everything downstream rides one cable. If you've seen rgthree's Context nodes, same instinct, just built around checkpoint-plus-prompt.

How it works

It loads the checkpoint, grabs the CLIP that came with it, and uses that CLIP to encode whatever you typed into the positive and negative boxes - exactly what a pair of CLIP Text Encode nodes would do, minus the nodes. MODEL, CLIP, VAE, and your two freshly-encoded conditionings all get packed into the pipe and sent out. The prompt fields are optional, so you can leave one blank, but the whole point is to fill them in and skip the separate encode step.

The inputs and outputs that matter

• ckpt_name (required) - the checkpoint picker, same list as the built-in loader.
• positive (optional, multiline text) - your positive prompt, typed as a string. The node encodes it for you.
• negative (optional, multiline text) - same, for the negative.

Note these are STRING inputs - actual text you type - which is what separates this node from its cousin Load Checkpoint (PIPE), where the conditioning slots take already-encoded CONDITIONING instead. Use this one when you want to write prompts at the loader; use that one when your conditioning comes from somewhere upstream.

Outputs: pipe (the BASIC_PIPE, complete with encoded prompts - wire it onward) and name_string, the checkpoint filename as text. Feed that into your save node's filename and your outputs come out self-labeled with the model.

Installing it

Pure Python, zero model downloads, nothing heavy - installs fast. Use ComfyUI Manager → Install Custom Nodes → search "antrobots ComfyUI Nodepack" → Install → restart, or clone it:

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

and restart. You'll find it under antrobots-ComfyUI-nodepack/flow-control.

Common issues

The prompts are encoded with this checkpoint's CLIP, at load time. That's usually what you want, but if you're on an SDXL-style model that expects dual text encoders or some special conditioning setup, a plain single-encode may not give you the fancier behavior a dedicated encode node would. For vanilla SD1.5/SDXL text prompts it's fine; for exotic conditioning, encode separately and use Load Checkpoint (PIPE) instead.

The usual pipe gotcha applies too: the pipe output is a BASIC_PIPE and only connects to nodes that understand one. Don't try to jam it into a stock KSampler - unpack it (Impact Pack's FromBasicPipe) or feed it this pack's KSampler with Pipe. And since this is a small one-person pack, the README's support model is simple: hit a real bug, open a GitHub issue - the author actively wants them.