Output Config: Load (lab)

Output Config: Load (lab) is the node that turns a config file into a handful of live, connectable outputs - and it's the one you'll actually use most of the time, because it picks the file through a browser dialog instead of making you type a backend path. It's the headline "dynamic outputs" node of the ComfyLab Pack: you load a file, and suddenly the node has one output per key, ready to be wired into your KSampler, your empty latent, your checkpoint loader.

How it works

Drop the node in, hit the Load output config button, and pick a JSON, JSON5, or YAML file. The backend validates it against a JSON Schema and, if it's good, the frontend creates an output for every top-level key. A config this small is enough:

{
  "seed": 618896223111156,
  "cfg": 7.5,
  "steps": 40
}

Three keys, three outputs. Wire them to their destinations and run. If the config has a problem, you get a detailed visual error report - popup or browser console, your choice via a dropdown - and the node won't generate outputs until you fix it.

Values can be bare (seed: 42) or a dict (cfg: { value: 7.5 }), and the dict form adds presentation options: label, shape, color_off/color_on (for "not connected" / "connected" states), and even type. That's cosmetic, but when you have a dozen outputs it genuinely helps you find the one you're looking for. JSON5 and YAML allow comments, so you can document your own configs; JSON works too if you prefer strictness.

Inputs and outputs

The schema shows a single data input of type HIDDEN - the frontend's internal handoff, nothing you set. Outputs don't show in the schema because they're generated per-file. An empty-looking node before you load anything is normal, not a bug.

Why you'd bother

The killer use case is test-case management. Keep a folder of config files - one per LoRA test, one per style run - and instead of remembering which seed/cfg/steps combo goes with which experiment, the workflow just loads the file. It also standardizes shared workflows: parameters live in a documented file, not scattered across widgets. The validation-at-load-time means a typo shows up the moment you load, not after ten minutes of queueing.

One honest word: if your config is two values, this is overkill - just type them into widgets. The node pays off when you've got a real collection of configs to manage.

Install

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/bugltd/ComfyLab-Pack.git
cd ComfyLab-Pack
pip install -r requirements.txt

Or ComfyUI Manager → ComfyLab Pack. Restart after installing. No model downloads; the heaviest dependency is opencv-python. If you need the backend variant - file located on the server while your browser is elsewhere - that's Output Config: Retrieve (backend) (lab), and the pack's docs say exactly when to reach for it (rarely, for most people).