Checkpoint Loader (LoraManager)

Functionally, this is the familiar "Load Checkpoint" node: pick a checkpoint file, get back a MODEL, CLIP and VAE to build the rest of your graph on. What makes it a LoRA Manager node rather than the stock one is that checkpoint management is a first-class feature of the pack - scanning, organizing, metadata, previews, disk cleanup - and this loader is the graph-side hook into that. Your checkpoints browse in the same UI as your LoRAs, with the same notes and preview-image treatment.

If you're already living in LoRA Manager for your LoRA collection, using its checkpoint loader keeps one consistent story: everything the manager knows about a checkpoint (the version, the trigger words for baked-in concepts, your notes) is attached to the same file you're loading here.

The honest take: if you don't use the manager's checkpoint features, the stock "Load Checkpoint" node does the identical job and there's no reason to switch. The payoff is entirely in the management layer - being able to browse, tag, add notes, download from CivitAI, and clean up disk space across your whole checkpoint collection in one UI, then load from that same organized library on the canvas.

How it works

You select a checkpoint from the dropdown and the node loads all three standard pieces from it: the diffusion model, the CLIP text encoder, and the VAE. It's the anchor node most SD1.5- and SDXL-lineage workflows start from - from here the MODEL and CLIP go through your LoRA loaders and into sampling, and the VAE handles decode at the end.

Inputs and outputs that matter

• ckpt_name (enum, required) - which checkpoint to load, listed from your checkpoints folder. (Empty until you have files there.)

Three outputs, the standard trio: MODEL (into LoRA loaders, then your sampler), CLIP (into your text encode or a LoRA loader that patches CLIP), and VAE (into VAE Decode).

How to install it

Via ComfyUI Manager: search lora-manager, Install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/willmiao/ComfyUI-Lora-Manager.git
cd ComfyUI-Lora-Manager
pip install -r requirements.txt

then restart. Light Python install; the checkpoints themselves you download separately (the manager can pull them from CivitAI if you add your API key in settings).

Common issues & troubleshooting

No checkpoints in the list. They need to be in ComfyUI's models/checkpoints folder (or a path the manager is configured to see). Drop them there and refresh the node.

A checkpoint has no baked VAE. Some checkpoints ship without a usable VAE and produce washed-out or artifacted decodes. If that happens, load a standalone VAE with a separate loader and use that instead of this node's VAE output - that's a checkpoint quirk, not a loader bug.

Match your LoRAs to it. Whatever architecture this checkpoint is - SD1.5, SDXL, Illustrious, Pony - your LoRAs have to match it. A LoRA for a different base will load without complaint and simply do nothing, which is the single most common "why isn't my LoRA working" cause.