Eric Qwen-Edit Apply LoRA:
Eric Qwen-Edit Apply LoRA loads a compatible adapter into a Qwen-Edit image-editing pipeline so you can influence the model’s results with a trained style, subject, or other learned effect. It supports standard LoRA loading and includes a fallback that attempts to handle certain adapter files with different key formats, including LoKR and LoHa. The node can reuse an adapter already loaded in the pipeline or clear loaded adapters before a fresh load. You can place multiple Apply LoRA nodes in a workflow, but whether their effects remain active together depends on how the underlying pipeline manages adapters; chaining nodes alone does not guarantee a combined effect.
Eric Qwen-Edit Apply LoRA Input Parameters:
pipeline
The required QWEN_EDIT_PIPELINE to modify. Connect a pipeline from your Qwen-Edit model-loading workflow. The node accesses the pipeline’s underlying model, applies or updates an adapter, and passes along the same pipeline data. This input has no selectable options, numeric range, or default value.
lora_name
The required dropdown for choosing an adapter file from ComfyUI’s configured loras folders, including their subfolders. It lists files ending in .safetensors, .bin, .pt, or .pth, plus a none option. Choose none to pass the pipeline through without loading an adapter, unless you provide a nonempty lora_path_override, which takes priority. The available filenames depend on the files in your installation; the source does not specify a fixed default selection.
weight
Controls the requested strength of the selected adapter. The default is 1.0, the minimum is -2.0, the maximum is 2.0, and the adjustment step is 0.05. As a starting point, 1.0 represents full strength and 0.5 represents half strength; values above 1.0 amplify the effect and may produce artifacts. A value of 0.0 requests no effect, but do not assume it reverses an adapter whose weight was merged directly into the model: changing that adapter’s strength requires reloading it.
lora_path_override
An optional text field for a custom adapter file path. Its default is an empty string. When you enter a nonblank path, the node uses that file instead of lora_name, even when the dropdown is set to none; the path must exist or the node raises an error. Leave it blank to use the dropdown. There is no numeric range or fixed list of options for this field.
keep_loaded
An optional boolean that controls whether previously loaded adapters are retained between runs. The default is True. With True, the node checks for an adapter with the same derived name and attempts to update its weight instead of loading it again. With False, it attempts to unload all previously loaded LoRA adapters before loading this one, which is useful when testing or swapping files but may take longer. In a chain, False can also remove adapters loaded by earlier nodes. Its options are True and False.
Eric Qwen-Edit Apply LoRA Output Parameters:
pipeline
The QWEN_EDIT_PIPELINE to connect to the next Qwen-Edit node. It is the same pipeline data supplied to the input, modified in place when an adapter is applied. After a successful load or update, the pipeline data also records the adapter’s path, requested weight, and filename under applied_loras; that record is not a separate output socket. If lora_name is none and no override path is supplied, the node returns the pipeline without applying an adapter.
Eric Qwen-Edit Apply LoRA Usage Tips:
- Start with
weightat its default of1.0, then adjust in0.05steps if the effect is too weak or too strong. Lower values such as0.5are useful when you want a subtler result. - Put adapter files in a configured ComfyUI
lorasfolder to select them fromlora_name. Uselora_path_overridewhen the file is elsewhere, and check the path carefully if loading fails. - Keep
keep_loadedset toTruewhen repeatedly rendering with the same adapter to avoid loading it from scratch on every run. Set it toFalsewhen you need to attempt a clean adapter state before testing another file. - When chaining Apply LoRA nodes, leave
keep_loadedatTrueon later nodes if you do not want them to unload earlier adapters. Check your rendered results to confirm which effects are active together. - If the log says an adapter was loaded by direct weight merge, reload it to change its effective strength; simply changing
weighton an already loaded direct-merge adapter will not rescale it.
Eric Qwen-Edit Apply LoRA Common Errors and Solutions:
LoRA file not found: <path or name>
- Explanation: The entered override path does not exist, or the file selected by
lora_namecan no longer be found in the configuredlorasfolders. - Solution: Check the filename and path, confirm the file is still present, and select it again or enter a valid override path.
[EricQwenEdit] Error loading LoRA: <details>
- Explanation: The pipeline could not load or apply the chosen adapter. The details after the colon come from the underlying loader and identify the specific problem.
- Solution: Read the full error details, confirm the file is an adapter intended for the loaded Qwen-Edit model, and try a known-compatible file to narrow down whether the issue is file-specific.
[EricQwenEdit] Unrecognised adapter format. First 5 keys: <keys>
- Explanation: Standard loading failed, and the fallback could not identify the file as a supported LoRA, LoKR, or LoHa adapter from its stored keys.
- Solution: Verify that you selected an adapter file rather than a full model or another file type. Obtain a compatible adapter export if the file is in an unsupported format.
Target modules ... not found or No modules were targeted
- Explanation: The adapter’s expected model components do not match what the loader found. The node attempts a key-format fallback for these errors, but loading may still fail.
- Solution: Confirm that the adapter was made for the Qwen-Edit model you loaded. If it should be compatible, try an export with keys supported by that model and loader.
[EricQwenEdit] NOTE: '<adapter>' was loaded via direct weight merge ... Changing weight requires reloading the LoRA.
- Explanation: This is a log note, not a loading failure. The adapter’s weight was incorporated when it loaded, so changing
weighton the already loaded adapter does not change its effective strength. - Solution: Reload the adapter at the desired weight, for example by running the node with
keep_loadedset toFalse. Be aware that this also attempts to unload all other adapters first.
[EricQwenEdit] set_adapters note: <details> (adapter may have been loaded via direct merge)
- Explanation: The node caught an error while asking the pipeline to set the adapter’s weight. Loading may have completed, but the requested strength is not confirmed by this message.
- Solution: Check the resulting image and the preceding load messages. If the strength is not what you expect, try reloading with
keep_loadedset toFalseand confirm the adapter is compatible with the pipeline.
KeyError: 'pipeline'
- Explanation: The supplied pipeline data does not contain the underlying
pipelineentry that this node expects. - Solution: Connect a valid
QWEN_EDIT_PIPELINEproduced by the appropriate Qwen-Edit loading workflow, rather than data assembled for a different pipeline type.
