CWK Wan2.2 LoRA Applier:
CWK_Wan22LoraApplier applies a stack of LoRAs one at a time using ComfyUI’s standard LoRA loader. This lets you combine the effects of several LoRAs in a defined order without applying each one through a separate node. It is intended for Wan 2.2 workflows, including graphs that process high-noise and low-noise models separately, and its use of the standard loader makes it suitable for looping graphs. If the stack is empty, the node passes its starting model and CLIP through unchanged.
CWK Wan2.2 LoRA Applier Input Parameters:
model
The required MODEL to which the stack’s model-side LoRA effects are applied. Connect the model for the branch you are processing, such as a high-noise or low-noise model. Each applicable LoRA is loaded onto the result of the previous one, so this input provides the starting point for the final model. The node defines no default value, minimum, maximum, or selectable options for this input.
clip
The required CLIP that accompanies the model. A LoRA in the stack can also affect CLIP, and the node passes the current CLIP through each successive LoRA load. Connect the CLIP associated with the branch you are processing; the resulting CLIP is especially relevant when you will use it for prompt encoding. The node defines no default value, minimum, maximum, or selectable options for this input.
lora_stack
The required LORA_STACK containing the LoRAs to apply. The source code expects each stack entry to contain a LoRA name, a model weight, and a CLIP weight, in that order. The weights are passed to ComfyUI’s standard loader; this node does not provide separate weight controls or define weight limits. Entries are processed in stack order, entries with an empty name are skipped, and an empty stack leaves the model and CLIP unchanged. The node defines no default value, minimum, maximum, or selectable options for this input.
CWK Wan2.2 LoRA Applier Output Parameters:
model
The resulting MODEL after the named LoRAs in the stack have been applied in order. Connect it to the next model-processing step in the same branch. If the stack is empty or every entry has an empty name, this output is the original model.
clip
The resulting CLIP after the stack’s applicable CLIP-side LoRA effects have been applied in order. Use it for downstream prompt encoding when that branch needs the LoRA-adjusted CLIP. If the stack is empty or every entry has an empty name, this output is the original CLIP.
CWK Wan2.2 LoRA Applier Usage Tips:
- For a high-noise branch, connect its model and LoRA stack to one instance of the node, then use the resulting model and CLIP where that branch needs them. For a low-noise branch, use a separate instance with its own model and stack; the source documentation notes that its CLIP output may be left unconnected when it is not needed.
- Check the order of entries in your stack. The node applies each named LoRA to the model and CLIP returned by the preceding load, so the stack order determines the loading sequence.
- Set model and CLIP weights in the node that builds your
LORA_STACK. This node reads those values from the stack; it has no controls for adjusting them independently. - When checking whether this node is affecting a result, inspect the execution log for lines beginning
[CWK LoraApplier] Applying. The node prints one such line for each entry it attempts to load, including that entry’s name and weights. An empty stack or an entry with an empty name produces no application line for that entry.
CWK Wan2.2 LoRA Applier Common Errors and Solutions:
Required input is missing: model or Required input is missing: clip
- Explanation: The node requires both a
MODELand aCLIPconnection. The exact validation wording can vary by ComfyUI version. - Solution: Connect both sockets from compatible upstream nodes, even if you only plan to use the model output of this branch.
Required input is missing: lora_stack
- Explanation: The stack socket is required, even though an empty stack passes the model and CLIP through unchanged. The exact validation wording can vary by ComfyUI version.
- Solution: Connect an upstream node that produces a
LORA_STACK. If you intend to apply no LoRAs, ensure that upstream node supplies an empty stack.
ValueError: not enough values to unpack or ValueError: too many values to unpack
- Explanation: The node expects each stack entry to have exactly three values: name, model weight, and CLIP weight. An entry with a different number of values cannot be read by the loop.
- Solution: Check the node creating the stack and ensure every entry follows the expected three-value structure.
TypeError: cannot unpack non-iterable ... object
- Explanation: A stack entry is not a three-part entry that the node can read. The object type shown in the message depends on the malformed entry.
- Solution: Use a stack-producing node that supplies entries containing a LoRA name, model weight, and CLIP weight.
A LoRA filename or loading error after [CWK LoraApplier] Applying
- Explanation: The node has passed the named entry to ComfyUI’s standard
LoraLoader, which could not load or apply it. The exact message depends on the loader and the underlying problem. - Solution: Check the printed LoRA name against the LoRAs available to your ComfyUI installation, then confirm that the selected LoRA is suitable for the model and CLIP in this branch.
No LoRA effect and no [CWK LoraApplier] Applying line
- Explanation: This is not necessarily an error. The node returns its inputs unchanged when the stack is empty and skips entries whose names are empty.
- Solution: Inspect the upstream stack and make sure it contains at least one entry with a nonempty LoRA name and the intended weights.
