Z-Image LoRA Stack Apply

The pack's main Merger node is convenient, but its LoRA slots are baked into the node itself. Once you're past two or three LoRAs - or you want to test how normalize compares to sqrt_scale without re-plugging cables - the stack pair is cleaner. Z-Image LoRA Stack Apply is the apply half of that pair: you hand it a lora_stack built with the pack's own Z-Image LoRA Stack nodes, plus your model and clip, and it applies the whole stack using the same six blend modes and the same target_strength logic as the Merger.

How it works

Same math under the hood, which is good news if you already learned the Merger. It pulls the strengths off the stack, rescales them by your chosen blend formula (default normalize, meaning sum of squares = target_strength²), then applies each LoRA sequentially through ComfyUI's standard loader. If the stack is empty or you never connected one, it passes model and clip straight through - handy when you're building a template that occasionally runs bare.

Inputs are compact: model, clip, lora_stack (type LORA_STACK), blend_mode, target_strength, and clip_strength_multiplier for scaling only the CLIP side. Outputs are model and clip, feeding your sampler as usual.

The one thing that burns people

LORA_STACK is a type name several packs share, but the payloads are not interchangeable. This pack's stack carries the actual loaded LoRA tensors, so a stack assembled by rgthree's LoRA Stack or Power Lora Loader won't drop into this node. Build the stack with this pack's Z-Image LoRA Stack node, apply it with this node, and you're fine. Cross-pack, don't.

Settings

Same advice as the Merger: on Z-Image Turbo start with normalize and target_strength at 0.7–0.9; use additive only as your baseline for comparison. The real value of this node is that you can flip the blend mode at a single point after the stack is built, instead of editing every LoRA individually.

Install

Nothing special - same as the rest of the pack, with no extra dependencies or model downloads:

ComfyUI Manager → search "Z-Image LoRA Merger" → install → restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/DanrisiUA/ComfyUI-ZImage-LoRA-Merger

Restart, and it appears in loaders/lora.

One honest note: if you already run rgthree's pack, its own stack handles most of this use case and it's battle-tested. This node earns its place if you specifically want the blend-mode normalization - which a plain stack won't give you - without dragging in a second QoL pack.