💊 CR LoRA Stack

Three LoRA slots, each with its own on/off switch and separate model/CLIP weights, bundled into one node instead of three separate loaders daisy-chained together. This node doesn't apply anything by itself - it builds a LORA_STACK that a downstream "apply" node (CR Apply LoRA Stack, elsewhere in the pack) actually bakes into your model and CLIP. Think of it as the recipe; the apply node is what cooks it.

How it works

Each of the three slots has switch_N (Off/On), lora_name_N, model_weight_N, and clip_weight_N. Flip a switch off and that slot is skipped entirely - the LoRA selection stays put, so you can re-enable it later without picking it again. An optional lora_stack input lets you chain another CR LoRA Stack node in, so three slots isn't actually a hard ceiling - daisy-chain as many of these as you need.

The output LORA_STACK is an ordered list; whatever's downstream (the apply node) walks through it in order and applies each active entry to the model and CLIP in turn.

Inputs and outputs that matter

• switch_N, lora_name_N, model_weight_N, clip_weight_N for N = 1, 2, 3.
• Optional lora_stack - chain another stack node's output in to go past three entries.
• Outputs: LORA_STACK, show_help.

How to install it

ComfyUI Manager: search "Comfyroll Studio", install, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/Suzie1/ComfyUI_Comfyroll_CustomNodes.git, restart. Your .safetensors LoRA files need to be sitting in models/loras already - the node just lists what's there.

Common issues

Worth an honest comparison, since it comes up: the pack most people reach for today when they want to stack several LoRAs with per-entry toggles is rgthree-comfy's Power Lora Loader, which does the same fixed-slots-with-switches idea but in a single node with effectively unlimited slots, and a right-click that can pull trigger words straight off the file. CR LoRA Stack predates that pattern and caps you at three per node (extendable by chaining), and it's still perfectly functional - old Comfyroll template workflows built around it work exactly as they did - but if you're starting a new workflow from scratch with no reason to stay in the Comfyroll ecosystem, Power Lora Loader is the more actively maintained version of the same idea.

The stacking-weights trap applies here same as anywhere: three LoRAs at 1.0 each is a lot of combined influence and tends to oversaturate or distort output. Pull weights down toward 0.5–0.8 per LoRA once you're combining more than one, and if two LoRAs are fighting for the same concept, lowering weights beats deleting one outright - you keep some of both effects rather than losing one entirely. And remember this node is a no-op by itself: if none of your changes here seem to affect generation, check that the stack is actually wired into a CR Apply LoRA Stack node downstream - this node alone doesn't touch your model.