LoRA Loader

If you've ever stacked four LoRAs in stock ComfyUI, you know the pain: a daisy chain of Load LoRA nodes, each with its own model in/out, and changing a weight means hunting through the chain. The Nynxz LoRA Loader collapses all of that into one node with a stack widget on the node itself - MODEL in, the stack, MODEL out. Each row has an on/off toggle, a searchable picker, and a strength, so you build the whole combo in one place instead of chaining single-LoRA loaders.

The stack widget

That's the pitch and it's a good one. The stack is a widget, not sockets: add rows, pick a LoRA per row (with sidecar preview thumbnails and bookmarks), toggle rows off when you want to A/B without deleting them, and set each strength. Bookmarks persist in a favorites.json beside the pack, so your go-to list survives restarts and even updates (it's gitignored and never shipped).

This variant is the MODEL-only loader - no CLIP. The README calls it the common case these days, and it mirrors ComfyUI's stock LoraLoaderModelOnly: plenty of modern checkpoints (Flux, Krea, the Qwen line) don't need their text encoder patched, so why pay for the extra wiring. If you do need to patch CLIP alongside the model, grab the sibling "LoRA Loader (CLIP)" instead.

Inputs, output, and when this beats the chain

Two inputs, one output: model in, the stack widget, patched model out. That's the whole graph footprint - UNETLoader → LoRA Loader → KSampler, with everything else living inside the widget. The stack's per-row strength applies to the model; if you need separate CLIP strength per row, that's the CLIP variant's territory.

One honest note from the author's own design: this loader keeps the stack private to the node. If you want the same stack applied in several places - say, the same combo on two different checkpoints - the headless pair (LoRA Picker → Apply LoRA) is the tool, since one picker can feed multiple Apply nodes. For a single workflow, this is the one you want.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/Nynxz/ComfyUI-NynxzExperimental

or search "Nynxz" in ComfyUI Manager, then restart. The pack itself has zero Python dependencies and downloads no models - your LoRAs come from wherever you already keep them. Like everything in this pack it's explicitly experimental: node ids and widget behavior can change between commits, so pin the version once your stack is dialed in.