🛡️ Easy LoRA Studio

A ComfyUI node in LoRA/Universal with 15 inputs and 5 outputs.

By Terpentinas·Created 6 months ago·Updated 3 months ago· 11

🛡️ Easy LoRA Studio

◄lora▾►

◄target_formatauto►

◄compression_modeoriginal►

◄target_rank128►

◄bake_custom_scale1.00►

◄save_triggerfalse►

◄filenameconverted_lora►

◄te_modeoriginal►

◄te_weight1.00►

◄precisionauto►

◄deviceauto►

◄save_folder/tmp/ComfyUI/models/loras►

CategoryLoRA/Universal

Inputs (15)

Name	Type	Default	Description
lora	COMBO		1 options: None
target_format	COMBO	auto	Auto‑select optimal format based on LoRA structure; otherwise choose manually. Standard (WebUI/Neo) = underscore + lora_down/up; Comfy Native = dot + lora_A/B; Forge‑Optimized = underscore + prefix mapping for Flux
compression_mode	COMBO	original	Original: no compression. Auto-fast: fast randomized SVD preserving 95% energy. Auto-full: full precision SVD preserving 95% energy (slower). Manual: use target rank slider.
target_rank	INT	1281–320	Target rank for manual compression (1‑320). Ignored if compression_mode is not manual.
bake_custom_scale	FLOAT	1.00-10–10	Multiply up‑weight tensors (lora_up.weight/lora_B.weight) by this factor before baking (e.g., 0.5 weakens the LoRA, 2.0 strengthens). Note: down‑weights are not scaled to avoid double‑strength effect.
save_trigger	BOOLEAN	false	—
filename	STRING	converted_lora	—
modelopt	MODEL		—
clipopt	CLIP		—
lora_dataopt	LORA		—
te_modeopt	COMBO	original	Control Text Encoder weights: original (keep as‑is), remove (strip TE keys), scale (multiply by te_weight)
te_weightopt	FLOAT	1.000–5	Multiplier for TE weights when te_mode is 'scale'
precisionopt	COMBO	auto	4 options: auto, float32, bfloat16, float16
deviceopt	COMBO	auto	3 options: auto, cuda, cpu
save_folderopt	STRING	/tmp/ComfyUI/models/loras	—