Load Checkpoint MXD

Load Checkpoint MXD is the core ComfyUI "Load Checkpoint" node wearing nicer clothes. Underneath it's doing exactly the same job - loading a diffusion checkpoint and handing you a MODEL, a CLIP and a VAE - but the dropdown comes with the pack's info-icon UI: CivitAI lookup, cached metadata, and your own local notes on the model. If you juggle more than a handful of checkpoints, that alone is worth the swap, because the core loader gives you a filename and nothing else to tell one SDXL merge from another.

How it works

It's a thin wrapper around ComfyUI's own loader. The node reads ckpt_name, resolves it through the standard checkpoints folder, and calls comfy.sd.load_checkpoint_guess_config - the same code path the core node uses - returning the model, CLIP and VAE in that order. The difference is entirely in the frontend. The MXD info-icon UI, which the pack also uses on its LoRA loader, is the same idea rgthree made famous: right-click / open the info panel and instead of a bare filename you get a lookup against CivitAI, cached metadata so it works offline on the second visit, and a notes field so you can annotate "this one's the good photorealism merge" next to the file. None of that changes the weights - it changes whether you can tell them apart.

Inputs

One input, ckpt_name, which is just the checkpoint dropdown from your ComfyUI/models/checkpoints folder (plus any folders you've auto-registered via the pack's model-storage setting). The three outputs - MODEL, CLIP, VAE - wire into the graph exactly like the core node's, so it's a drop-in replacement: swap the node, keep every downstream wire.

Installing

ComfyUI Manager → search "Maxed Out" → install, then restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/Maxed-Out-99/ComfyUI-MaxedOut.git

Then restart ComfyUI. No pip dependencies and no extra model files - it loads whatever checkpoints you already have.

Common issues

Because it's a wrapper, the usual checkpoint-loader gotchas apply: a checkpoint that fails to load in core will fail here too, and VAE/CLIP guess-config quirks (e.g. a merged model that misdetects its architecture) are unchanged. The real gotcha is the notes being local - the info panel caches metadata and stores your annotations per machine, so they don't travel with a shared workflow. And if your ckpt_name dropdown is missing files you know exist, check the model-storage auto-registration setting (Settings → MXD → Model Storage); folders outside ComfyUI/models need that root configured before startup for them to appear.