CivitaiManager - Multi LoRA Loader

This is the node at the end of CivitaiManager's "find it, download it, use it" chain - the step where all those LoRAs you browsed and queued actually land in your workflow. Instead of chaining three, four, or seven individual LoraLoader nodes down the left edge of your graph, you get one node that holds a whole list: each entry with its own on/off toggle and strength, reorderable, and fed by a popup LoRA manager that can search Civitai and download-and-apply without you leaving the canvas. If you live inside CivitaiManager, this replaces the loader spaghetti outright.

The name is honest: it's CivitaiManager - Multi LoRA Loader, living under the CivitaiManager/Loaders category. It's new - the pack only hit 1.0.0 in late July 2026 - so don't expect a decade of community lore around it. What it does is straightforward and well-built.

How it works

Under the hood it's a loop, not magic. The node parses its lora_list_json into an ordered list of entries (each carrying a filename, a strength, an enabled flag, and optionally a storage-root id), then walks them in order, feeding the output of one LoRA into the next: load_lora_for_models chained through every entry. Order is order - LoRAs are patches on top of patches, and stacking is powerful but unpredictable, so the sequence genuinely matters (a character LoRA then a style LoRA behaves differently from the reverse). Two behaviors worth knowing: missing files are skipped, not fatal - the list keeps going and the skip is logged to your ComfyUI console - and each entry applies only a model strength; there's no separate clip-strength knob per LoRA in this node. That's a reasonable simplification, not a bug.

The inputs that matter

Only two, and only one of them is yours to manage:

model (MODEL) - whatever you loaded from your Checkpoint/UNet loader. Wire it in.
lora_list_json (STRING) - trust the tooltip: "Managed by the CivitaiManager LoRA selector." You don't hand-type this JSON. Click the node's "Open LoRA manager" button, pick LoRAs, tweak their strengths (0.5–0.8 is a common sweet spot - 1.0 is often too strong), reorder, toggle, done.

The single model output plugs straight into your sampler or into whatever else modifies the model on its way to the KSampler.

Installing it

You're really installing the whole pack, since this node is one piece of it:

cd ComfyUI/custom_nodes
git clone https://github.com/nregret/CivitaiManager

Then restart ComfyUI. Or use ComfyUI Manager and search "CivitaiManager". Either way there's no pip install and no heavy dependencies - the shipped project declares zero Python dependencies, so nothing extra to break. The LoRA selector popup is part of the pack's frontend, so keep the js/ folder intact.

Where people get burned

Mostly it's the usual LoRA traps, not this node's fault. A LoRA trained on Illustrious won't do anything sensible on Flux - match the base model or you'll get mush and blame the wrong tool. If a LoRA you're sure you downloaded doesn't apply, check the ComfyUI console for a line like [CivitaiManager] LoRA not found, skipped: <name> - the node skips silently-ish by design, and it's logging exactly where it looked. And if lora_list_json reads [], the list is genuinely empty; hit the selector button. Once you've got the chain right, this is the one node I'd actually reach for over a pile of single LoRA loaders - it's the rare manager feature that pays for itself in graph cleanup alone.