LoRA Apply (MRLN)

Apply the LoRA blocks a Prompt Template drew onto MODEL and CLIP. Wire the Prompt Template's 'loras' output into this node between your model/clip loaders and the sampler: every drawn LoRA block (a section item carrying lora + strength metadata) is loaded with its authored strengths — the deterministic draw decides the LoRA stack, no tag-parsing loader needed. With no LoRA blocks drawn, model and clip pass through unchanged.

By master-merlin·Created 16 days ago·Updated 7 days ago· 2

LoRA Apply (MRLN)

model
clip

model
clip
report

◄loras—►

◄on_missingerror►

◄on_mismatchwarn►

CategoryMRLN/prompt

Inputs (5)

Name	Type	Default	Description
model	MODEL		Model to apply the drawn LoRA stack to.
clip	CLIP		CLIP to apply the drawn LoRA stack to.
loras	STRING		The 'loras' output of a Prompt Template (MRLN) node — a JSON list of the drawn LoRA blocks.
on_missingopt	COMBO	error	A drawn LoRA file this machine does not have: 'error' stops the run and names the file (safe default); 'skip' renders without it and logs a warning — the trigger words stay in the prompt, so the image just loses that LoRA's influence; 'download' fetches it from Civitai by the AIR urn stored on the item (SHA256-verified, blocks the run for as long as the download takes) and re-points the item at the file. Use 'download' for shared workflows that should heal themselves without opening the Composer.
on_mismatchopt	COMBO	warn	A LoRA trained for a different base model than the connected one (a FLUX LoRA on an SDXL or KREA checkpoint) loads without erroring but quietly degrades the image. 'warn' logs the mismatch and applies it anyway; 'skip' leaves that LoRA out; 'error' stops the run; 'ignore' disables the check. Only LoRA items that declare a family (data.base, or the ecosystem segment of their Civitai AIR) can be checked.

Outputs (3)

Name	Type	Description
model	MODEL	The model with every drawn LoRA applied at its authored strength.
clip	CLIP	The CLIP with every drawn LoRA applied at its authored strength.
report	STRING	What actually happened, line by line: each LoRA applied with its strengths, plus any base-model mismatch, missing file, skip or download. Wire it into Show Text (MRLN) to see it in the graph — silent degradation is the failure mode this catches.