Krea2 Merge • Apply LoRA

Let's be honest about this one up front: Krea2 Merge • Apply LoRA is basically the stock Load LoRA node wearing a different hat. It takes a model, a LoRA file, and a strength, and hands you back a model with that LoRA applied. ComfyUI has shipped that exact node since Krea 2 support landed natively. So why would you reach for this copy?

Because you're already living in the Krea2 Merge workflow. The pack's whole job is merging two to four Krea 2 LoRAs into one file, and this node is the pack's answer to "how do I try each LoRA before I commit to merging it?" It puts a strength slider on any Krea 2 (or Kohya) LoRA in the same graph you're merging in, so you can audition sources - or the freshly merged file itself - without leaving the category. It also carries the pack's folder-filter dropdown, and unlike the stock loader it remembers your filter across workflow loads, which is a nicer quality of life than it sounds.

How it works

Under the hood it does the unglamorous but standard thing: it resolves the file from ComfyUI/models/loras, loads the .safetensors (or .pt/.ckpt) into a state dict, and hands it to ComfyUI's own comfy.sd.load_lora_for_models. That's the same code path stock loaders use, and it's why this node handles both the lora_A/lora_B PEFT keys Krea 2 LoRAs ship with and classic Kohya lora_down/lora_up keys without caring. Krea 2 LoRAs tend to be trained at rank 16 with alpha 16 - the ai-toolkit default - and this loader treats them like any other LoRA.

The inputs that matter

• model - a real MODEL, straight from a UNETLoader. This is the node's one big difference from the pack's plain Load LoRA.
• lora_name - the file picker, filtered by the category_filter folder dropdown above it.
• strength_model - a float from -2.0 to 2.0, default 1.0. Negative values subtract rather than add, which is a real trick for softening a style that's baked in too hard.

The output is a single model that wires straight into your KSampler or sampler chain - anything that accepts a MODEL.

Install

Same as the rest of the pack: grab it in ComfyUI Manager by searching Krea2 Merge (the pack is published on the Comfy Registry), or clone it in:

cd ComfyUI/custom_nodes
git clone https://github.com/cyberdeliaAI/comfyui-krea2-merge

Then restart ComfyUI. That's the whole install - the only real dependency is safetensors, which every normal ComfyUI install already has, and the pack never downloads or modifies the Krea 2 base model.

Gotchas

If the node complains the file isn't found, the LoRA isn't under models/loras (or the folder filter is hiding it - flip category_filter to "All"). And don't expect this node to feed the merger: Apply LoRA produces a working model, not a state dict, so the merge pipeline wants the pack's Load LoRA node instead. Use the right tool per socket and this one is quietly, boringly reliable.