Meld Unified Loader

In a stock ComfyUI workflow the model, the prompts, and the KSampler's settings all live in separate nodes, and keeping them in sync across a big graph is exactly the kind of bookkeeping people burn weekends on. Meld Unified Loader is the Meld pack's answer: one node that loads the checkpoint and holds the prompt and sampling parameters, then hands you a single base_settings dict you can reuse or unpack anywhere.

It comes from the same family as Meld Unified Flux Loader, but this one is the general-purpose sibling - it loads a plain checkpoint (SD 1.5, SDXL, whatever sits in models/checkpoints) the way Load Checkpoint does, just with all the generation settings bundled in.

How it works

Under the hood it calls comfy.sd.load_checkpoint_guess_config with your checkpoint - the same load path the stock checkpoint loader uses, including the CLIP and VAE extraction. The difference is everything else on the node. Alongside model_name you get positive, negative, seed, steps, cfg, sampler_name, scheduler, width, and height all in one place, and every one of those values gets packed into a base_settings dict output.

The payoff is that you can treat that dict as a single "what made this image" blob. Wire it to Meld Settings Unpacker and it becomes real seed/steps/cfg wires you can route into a KSampler, or into a display node, or back into another loader.

The inputs you'll actually touch

• model_name - the checkpoint, from models/checkpoints.
• positive / negative - multiline prompt fields, so you can skip the separate text node for simple workflows.
• cfg - defaults to 8.0, which is a sensible starting point for SD-family models.
• width / height - default 512×512; bump to 1024 for SDXL-class models.

Outputs: MODEL, CLIP, VAE, positive, negative, and base_settings. Wire MODEL/CLIP/VAE into a KSampler as usual.

Where it shines: the Image Manager loop

This is where the node stops being "a convenience loader" and starts being a workflow. If you use Meld's Image Manager, open an image's menu (the ⋮ on a card), pick "Add Unified Loader," and ComfyUI drops a Meld Unified Loader into your graph pre-filled with that image's prompt, model, seed, and settings. Reproducing or tweaking an old generation becomes a two-click operation instead of reconstructing every value by hand.

Installing it

Same pack, same install for every Meld node:

• ComfyUI Manager → search "Meld" → install, or
• comfy node install HappyOnigiri/ComfyUI-Meld, or
• cd ComfyUI/custom_nodes && git clone https://github.com/HappyOnigiri/ComfyUI-Meld.git && cd ComfyUI-Meld && pip install -r requirements.txt then restart ComfyUI.

The pack's requirements pull in aiohttp, imagehash, scipy, and PyWavelets on top of what ComfyUI already ships; nothing exotic here.

Worth knowing

The base_settings dict only carries what you set on the node - it doesn't chase every value in the graph, so connect the sampler settings through the loader (or the Unpacker) if you want them tracked. And as with the rest of the pack, the author is Japanese and handles GitHub issues in machine-translated English; a screenshot plus a short description beats a long message every time.