LoRA Iterator — ComfyUI Custom Node

A ComfyUI node that automatically steps through every LoRA in a directory (or subdirectory) on each generation run, so you can test all your LoRAs without touching the graph between runs.

https://www.theoath.studio/projects/comfy-lora-iterator

Install

cp ComfyUI/custom_nodes/
git clone git@github.com:OATH-Studio/comfy-LoRA-iterator.git
# restart ComfyUI

The node appears in the node browser under loaders/lora → LoRA Iterator.

Inputs

Input	Type	Description
`model`	MODEL	Base model passthrough
`clip`	CLIP	Base CLIP passthrough
`directory`	Dropdown	`[All]` or a specific subdirectory of your loras folder
`lora_name`	Dropdown	The currently selected LoRA (relative path)
`strength_model`	Float slider	LoRA strength applied to the diffusion model
`strength_clip`	Float slider	LoRA strength applied to the CLIP encoder
`control_after_generate`	Dropdown	How to advance the LoRA selection after each run

Outputs

Output	Type	Description
`model`	MODEL	Patched model with LoRA applied
`clip`	CLIP	Patched CLIP with LoRA applied
`lora_name`	STRING	Name of the LoRA used this run (useful for filename nodes)
`current_index`	INT	Zero-based index of the current LoRA
`total_loras`	INT	Total LoRAs in the selected directory

control_after_generate

This is the key feature. After each generation completes, the node advances its internal LoRA selection according to the chosen mode:

Mode	Behaviour
`fixed`	Always uses the same LoRA — nothing changes between runs
`increment`	Moves forward one LoRA each run. Wraps back to 0 after the last one
`decrement`	Moves backward one LoRA each run. Wraps to the last one after index 0
`randomize`	Picks a random LoRA from the list each run

Example: You have 8 LoRAs in the zimage directory. Set directory to zimage and control_after_generate to increment. Queue 8 runs — each generation automatically uses the next LoRA with zero manual intervention.

Directory organisation

The directory dropdown shows [All] plus every immediate subdirectory inside your ComfyUI loras folder. This lets you group LoRAs by the model family they were trained on:

ComfyUI/models/loras/
    zimage/               ← appears as "zimage" in dropdown
        AnimeMix.safetensors
        Realism.safetensors
    sdxl/                 ← appears as "sdxl" in dropdown
        portrait.safetensors
    my_lora.safetensors   ← included when "[All]" is selected

Selecting zimage means only LoRAs trained for that model family are shown and iterated — preventing mismatched LoRA/model combinations.

Multiple instances

Each node instance tracks its own LoRA position independently using ComfyUI's internal unique_id. Two LoRA Iterator nodes in the same graph increment through their respective directories separately.

Typical workflow

CheckpointLoader → LoRA Iterator → KSampler → VAEDecode → SaveImage
                        ↑
                  set control_after_generate = increment
                  queue N runs (one per LoRA)

Wire lora_name output into a filename prefix node to embed the LoRA name in every saved image automatically.