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 > OmniNodes > LoKr Loader (Custom) 🧩

ComfyUI Node: LoKr Loader (Custom) 🧩

Class Name

LoKrLoaderNode

Category
TensorVizion/Model
Author
tensorvizion (Account age: 0days)
Extension
OmniNodes
Latest Updated
2026-10-28
Github Stars
0.0K

How to Install OmniNodes

Install this extension via the ComfyUI Manager by searching for OmniNodes
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter OmniNodes 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

LoKr Loader (Custom) 🧩 Description

Loads compatible LoKr patches into a cloned model at adjustable strength and reports applied layers.

LoKr Loader (Custom) 🧩:

LoKrLoaderNode applies a LoKr adaptation to a model, letting you change how the model behaves without replacing its underlying weights. LoKr builds each supported weight update from a Kronecker product: it combines two smaller matrices into an update shaped like a larger model weight. This differs from the multiplication used by LoRA and the element-wise combination used by LoHa. The node handles LoKr factors stored either as complete matrices or as paired low-rank matrices, and reports how many layers it could apply. It supports layers whose resolved factors are two-dimensional; convolutional layers with four-dimensional factors are not supported.

LoKr Loader (Custom) 🧩 Input Parameters:

model

The MODEL to adapt. Connect the model you intend to use for generation. The node clones it before adding any compatible LoKr patches, so its normal result is an adapted model that you can pass to later nodes. Compatibility matters: a LoKr layer must map to a weight in this model and produce an update of exactly the same shape. There is no default, minimum, maximum, or selectable option for this required connection.

lokr_name

The name of the file to load from the available loras files. This is a required dropdown selection, with options supplied by the files currently listed in that location; the source does not define a fixed list or a default choice. Choose a file containing complete LoKr factor groups. A file may be selectable yet produce no applied layers if it is not a compatible LoKr file, its layers do not match the connected model, or it uses unsupported factor shapes.

strength

The amount by which the node scales the LoKr patches it adds. The default is 1.0, the minimum is -5.0, the maximum is 5.0, and the control changes in steps of 0.01. Use 1.0 for the file’s full applied effect, a value between 0 and 1 for a weaker effect, or 0 for no patch effect. Negative values reverse the direction of the applied update, while values above 1.0 amplify it; these settings may produce results quite different from the effect you expect.

LoKr Loader (Custom) 🧩 Output Parameters:

model

The MODEL to connect to the rest of your workflow. When the file contains complete, compatible, supported layers, this output is a clone of the incoming model with their LoKr patches added at the chosen strength. If the file cannot be found or contains no complete LoKr layer groups, the node returns the original model instead. A model output does not, by itself, confirm that any layers were applied; check the summary.

summary

A STRING describing the load result. For a processed file, it lists the file name, layers applied, incomplete groups skipped, unsupported convolutional or four-dimensional factors skipped, layers skipped because their shape did not match or their model key was not found, and the selected strength. If the file cannot be found or has no complete LoKr groups, this output instead contains a short diagnostic message. Use it to distinguish a successful load from a file that had little or no effect.

LoKr Loader (Custom) 🧩 Usage Tips:

  • Start with strength at its default of 1.0, then adjust it in small steps while comparing results. Lower positive values can make an adaptation subtler; check the summary to confirm that layers were actually applied before judging the visual effect.
  • Choose a LoKr file intended for the model you connected. If the summary reports layers skipped for shape mismatch or a missing model key, check the model and file pairing; increasing strength will not make skipped layers apply.
  • Check the file with LyCORIS Format Inspector when the summary says no complete LoKr layer groups were found. The node expects both factors for each layer, either as dense lokr_w1 and lokr_w2 values or as complete _a and _b pairs for the relevant factor.
  • Look at all the counts in the summary, not only “Layers applied.” A file can apply some layers while skipping others, so its effect may be partial.
  • For this node, favor files with two-dimensional factors. Four-dimensional convolutional factors are skipped rather than applied.

LoKr Loader (Custom) 🧩 Common Errors and Solutions:

File not found: <lokr_name>

  • Explanation: The node could not resolve the selected name to a file in its loras location. It returns the original model with this message in summary.
  • Solution: Confirm that the file is available in the location used for LoRA files, refresh the file list if needed, and select an available name.

No complete LoKr layer groups found in <lokr_name> — confirm this is actually a LoKr file with LyCORIS Format Inspector first.

  • Explanation: The file loaded, but the node found no layer group containing both required LoKr factors. It returns the original model rather than applying an update.
  • Solution: Inspect the file format and select a LoKr file with complete factor data. If the file was damaged or exported incorrectly, obtain or export a valid copy.

Layers skipped (incomplete key set): <count>

  • Explanation: This is a line in the summary, not necessarily a fatal error. The node found one or more layer groups missing a complete first or second factor, and skipped those groups. Other complete groups may still apply.
  • Solution: Check whether the file is a complete LoKr export. Review “Layers applied” to see whether the remaining layers loaded.

Layers skipped (conv/4D factors — not yet supported): <count>

  • Explanation: One or more complete groups could not be used because their resolved factors were not both two-dimensional. The node does not implement convolutional or four-dimensional LoKr factors.
  • Solution: Use a LoKr file with supported two-dimensional factors for this node. Check “Layers applied” to determine whether other groups were usable.

Layers skipped (shape mismatch / key not found in model): <count>

  • Explanation: One or more groups could not be matched to a model weight of the required shape. This count can also include groups whose factor calculation failed for a reason other than the explicitly handled unsupported-factor case.
  • Solution: Confirm that the LoKr file was made for the connected model and review the other summary counts. If the models match but this count remains high, check the file’s layer names and factor dimensions.

Layers applied: 0

  • Explanation: No LoKr patches were added. This can occur even when the file contains complete groups, if none map to compatible, supported weights in the connected model.
  • Solution: Read the skipped-layer counts to identify the likely cause, then try a compatible model or a different LoKr file. Changing strength alone cannot fix an unapplied layer.

File loading error from load_torch_file

  • Explanation: The node calls the file loader without handling its exceptions, so an unreadable, invalid, or otherwise unloadable file may stop execution instead of producing the normal summary.
  • Solution: Check that the selected file exists and is a valid, readable model-adaptation file. Replace a damaged file and try again.

Patch application error from add_patches

  • Explanation: The node does not catch an exception raised while adding patches to the cloned model, so execution may stop without returning its summary.
  • Solution: Check that the incoming connection supplies a compatible MODEL and try a LoKr file intended for that model. The exact cause depends on the error reported by your workflow.

LoKr Loader (Custom) 🧩 Related Nodes

Go back to the extension to check out more related nodes.
OmniNodes
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.

LoKr Loader (Custom) 🧩