Krea2 controlnet lora loader

This is a small, single-purpose node with a specific job: it loads a Krea2 depth controlnet-lora file and outputs a MODEL_PATCH. That patch is what you feed into this pack's DiffSynth ControlNet node to get depth-map structural control in your workflow. If you googled this node, you're almost certainly building a depth-guided setup and hit the "how do I load the control-lora" step. This is that step.

A bit of context on the name. Krea 2 is Krea's open 12B image model, released June 2026, well regarded for stylized and cinematic work and for shipping a genuinely trainable base. A "depth control-lora" is a control adapter that lets a depth map dictate the structure of the generated image, the classic ControlNet trick where the prompt handles what and the depth map handles where. This node's whole role is to read that control-lora off disk and package it as a patch the rest of the chain can use.

How it works

The node is deliberately thin. You point it at a Krea2 depth controlnet-lora file (something like krea2-depth-control-lora.safetensors), and it emits a MODEL_PATCH. It doesn't touch your image or your model directly, it just loads the control weights and wraps them so a downstream node can apply them. Think of it as the "loader" half of a two-part setup: this node loads, and the DiffSynth ControlNet node applies.

The inputs and outputs that matter

There's almost nothing to configure, which is the point:

• name (required) - the dropdown of your available Krea2 depth control-lora files. Pick the one you want to use. That's the entire input surface.

Output is a single MODEL_PATCH. You wire it into the model_patch input of the NunchakuQI&ZITDiffsynthControlnet node, which also takes your Nunchaku model, a VAE, and your control image, then applies the depth guidance and hands a conditioned model to your sampler. On its own this node does nothing visible; it only makes sense as the front end of that ControlNet chain.

Installing it

It ships with the pack. In ComfyUI Manager, search for it, install, and restart. Or from a terminal:

cd ComfyUI/custom_nodes
git clone https://github.com/ussoewwin/ComfyUI-QwenImageLoraLoader

then restart. The node itself is lightweight, but the workflow it belongs to isn't self-contained: you'll want the pack's DiffSynth ControlNet node to consume the patch, a working official ComfyUI-nunchaku setup for the model side, and, separately, the actual Krea2 depth control-lora file downloaded into the folder the node reads from. Python 3.11+.

Where people get burned

• Empty dropdown. If name shows nothing, the node can't see any control-lora files. Put the Krea2 depth control-lora in the location the node scans and restart ComfyUI so it re-indexes.
• Nothing happens on its own. This is a loader, not an applier. The output is a MODEL_PATCH that has to go into the DiffSynth ControlNet node's model_patch input. Loading the file and stopping there gets you no control at all; wire the patch through.
• Match the depth map. Depth control only helps if the control image is an actual depth map of the composition you want. Feed the DiffSynth ControlNet node a real depth pass, not a raw photo, and tune that node's strength if the guidance is too strong or too weak.