Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > Krea 2 Reference > Krea 2 Reference LoRA (layered)

ComfyUI Node: Krea 2 Reference LoRA (layered)

Class Name

Krea2ReferenceLoraLoader

Category
loaders
Author
konokoaz (Account age: 1342days)
Extension
Krea 2 Reference
Latest Updated
2026-07-15
Github Stars
0.04K

How to Install Krea 2 Reference

Install this extension via the ComfyUI Manager by searching for Krea 2 Reference
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter Krea 2 Reference in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

Krea 2 Reference LoRA (layered) Description

Loads a Krea 2 reference LoRA with separate text-fusion and DiT body strength controls.

Krea 2 Reference LoRA (layered):

This node loads a reference LoRA and applies it to a Krea 2 model with separate strength controls for the text-fusion layers and the DiT body layers. This separation lets you control two different effects: the text-fusion and txt_in layers primarily influence how the model addresses references, reduces unwanted visual blending, and transfers style, while the DiT body layers add stronger scene or image-content adherence but may cause the generated result to drift away from the original model’s identity. By adjusting these layer groups independently, you can preserve the base model’s rendering behavior while still following the visual information contained in a reference LoRA, or increase scene adherence when the reference needs to have a stronger influence.

Krea 2 Reference LoRA (layered) Input Parameters:

model

The Krea 2 MODEL that receives the reference LoRA adjustments. The node creates an updated model by loading the selected LoRA weights and applying the configured strengths to the relevant layer groups. The model must be compatible with the LoRA, especially because the node separates keys beginning with transformer.text_fusion. and transformer.txt_in. from all other LoRA keys; incompatible model or LoRA architectures may fail during loading or produce unsuitable results. This parameter has no numeric range or default value defined by the node and must be connected to a model-producing node.

lora_name

The LoRA file to load from the configured LoRA folder. The available options are provided by the installation's LoRA file list, so you can select one of the LoRA files recognized by the workflow environment. The selected file is loaded with safe tensor-file loading enabled and then divided into two groups: keys beginning with transformer.text_fusion. or transformer.txt_in. are controlled by strength_text_fusion, while all remaining keys are controlled by strength_dit_body. The node does not define a fixed default, minimum, or maximum for this parameter; its available options depend on the LoRA files installed in the configured LoRA directory.

strength_text_fusion

Controls the strength applied to LoRA weights associated with the transformer.text_fusion. and transformer.txt_in. layer groups. These layers are intended to influence reference addressing, anti-bleed behavior, and style transfer. The default is 1.0, with a minimum of -2.0 and a maximum of 2.0; the adjustment step is 0.05. A value of 1.0 applies the text-fusion portion at its normal strength, while 0.0 skips this group entirely. Positive values increase the reference and style influence, and negative values apply the LoRA in the opposite direction, which can produce significantly different or unpredictable visual effects depending on the LoRA. When the DiT body strength remains at 0.0, using 1.0 here is intended to preserve the base model's rendering behavior while still allowing the workflow to follow reference information.

strength_dit_body

Controls the strength applied to all LoRA weights that are not associated with the transformer.text_fusion. or transformer.txt_in. prefixes. This group primarily affects the DiT body and is used for additional scene adherence. The default is 0.0, with a minimum of -2.0 and a maximum of 2.0; the adjustment step is 0.05. A value of 0.0 skips this group, while positive values make the reference LoRA influence the scene more strongly. The node documentation indicates that increasing this value above approximately 0.3 can cause identity drift, meaning the generated subject or overall appearance may move farther from the base model's identity. Negative values reverse the LoRA contribution and should be used cautiously because their visual effect depends on the specific LoRA.

Krea 2 Reference LoRA (layered) Output Parameters:

model

The modified MODEL containing the selected reference LoRA's applied adjustments. The output can be connected to downstream sampling or generation nodes in the same way as the original model. The text-fusion portion is applied using strength_text_fusion, and the remaining LoRA weights are applied using strength_dit_body; any group with no matching weights or a strength of exactly 0.0 is skipped. The output is still a model object rather than an image, so you must pass it to a sampler or another model-processing node to generate images.

Krea 2 Reference LoRA (layered) Usage Tips:

  • Start with strength_text_fusion at 1.0 and strength_dit_body at 0.0 when you want the base Krea 2 model to retain its normal rendering character while following the reference and transferring its style.
  • Increase strength_dit_body gradually in small steps, such as 0.05 or 0.1, when the generated scene does not adhere closely enough to the reference. Watch for identity drift, especially once the value approaches or exceeds approximately 0.3.
  • Keep strength_text_fusion and strength_dit_body independent while testing. This makes it easier to determine whether a change in reference behavior comes from addressing and style transfer or from stronger DiT body influence.
  • Use a value of 0.0 for either strength when you want to disable that layer group. This can be useful for comparing text-fusion-only behavior with the combined effect of both layer groups.
  • Treat negative strength values as experimental settings. They reverse the LoRA contribution rather than simply reducing it, so they may weaken or alter the reference effect in ways that vary between LoRAs.
  • Select a LoRA specifically prepared for the Krea 2 model architecture whenever possible. A LoRA made for a different architecture may not contain matching layer names or may not load correctly.
  • If changing the strength values appears to have no effect, check whether the selected LoRA contains keys for the corresponding group. The node skips a group when no matching keys are found, and it also skips any group whose strength is 0.0.
  • Keep the workflow's model and LoRA choices consistent during comparisons. Changing both the base model and LoRA at the same time makes it difficult to determine which change affected the generated image.

Krea 2 Reference LoRA (layered) Common Errors and Solutions:

FileNotFoundError or a file-not-found error for the selected LoRA

  • Explanation: The selected LoRA file cannot be found at the path supplied by the configured LoRA folder, or the file was moved, renamed, or removed after the node's file list was created.
  • Solution: Confirm that the LoRA file exists in the configured LoRA directory, restore or reinstall the file if necessary, and refresh the workflow environment so the lora_name options are rebuilt. Then select the file again.

ValueError from get_full_path_or_raise("loras", lora_name)

  • Explanation: The selected name does not resolve to a valid file in the configured LoRA locations. This can occur when a workflow references a LoRA that is not installed or when the file path is no longer valid.
  • Solution: Install or copy the required LoRA into a recognized LoRA folder, verify that the filename is correct, and choose the installed file from the node's lora_name list instead of relying on an outdated workflow selection.

An error while loading the LoRA file with safe loading enabled

  • Explanation: The selected file may be corrupt, incomplete, in an unsupported format, or not a valid PyTorch LoRA state-dictionary file. The node loads the file with safe loading enabled, so files that cannot be safely parsed are rejected.
  • Solution: Download or export the LoRA again from a trusted source, verify that the file is complete, and use a supported LoRA file format. Do not rename an unrelated model file to make it appear to be a LoRA.

Missing or incompatible model keys during LoRA application

  • Explanation: The LoRA may have been created for a different model architecture or may not contain layer names that match the connected Krea 2 model. The node classifies keys using the transformer.text_fusion. and transformer.txt_in. prefixes and applies the resulting groups through the model LoRA loader; incompatible keys can therefore fail to apply or produce an error.
  • Solution: Connect the Krea 2 model expected by the reference LoRA and select a LoRA designed for that architecture. If the error persists, try another known-compatible LoRA rather than increasing the strength values.

The node completes, but the LoRA appears to have no effect

  • Explanation: A layer group is skipped when it contains no matching keys or when its strength is exactly 0.0. For example, strength_dit_body = 0.0 intentionally prevents the non-text-fusion group from being applied, and a LoRA without matching text-fusion keys will not respond to strength_text_fusion.
  • Solution: Check both strength values and test the groups separately. Use a nonzero value for the group you want to test, confirm that the selected LoRA is intended for Krea 2 reference use, and compare the output with the original model.

The result follows the scene but the subject identity changes

  • Explanation: The DiT body contribution is too strong for the desired result. The node's documented behavior warns that identity can drift when strength_dit_body rises above approximately 0.3.
  • Solution: Reduce strength_dit_body, return it to 0.0, and rely more on strength_text_fusion for reference addressing and style transfer. Increase the DiT body value only in small increments when additional scene adherence is necessary.

The result has excessive style transfer or reference influence

  • Explanation: strength_text_fusion is applying a stronger contribution from the text-fusion and txt_in layers than the image requires. Positive values increase the influence of reference addressing and style-related behavior.
  • Solution: Lower strength_text_fusion gradually, compare the result at values such as 0.75, 0.5, or 0.0, and keep strength_dit_body low while isolating the cause.

The output is unexpectedly distorted when using a negative strength

  • Explanation: Negative strengths reverse the LoRA contribution. This is not the same as turning the effect down and can produce unusual results depending on the weights in the selected LoRA.
  • Solution: Return the affected strength to 0.0 or a small positive value, confirm that the workflow behaves as expected, and use negative values only when the specific LoRA has been tested for that purpose.

Krea 2 Reference LoRA (layered) Related Nodes

Go back to the extension to check out more related nodes.
Krea 2 Reference
RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

Krea 2 Reference LoRA (layered)