Multi LoRA Loader

A powerful multi-LoRA management node for ComfyUI with optional LTX2 layer-specific strength control, drag-and-drop reordering, and rgthree LoRA info integration.

Renamed from Power LTX LoRA Loader Extra. If you have existing workflows using the old node, see Migration below.

Features

• Multi-LoRA Support: Load and apply multiple LoRAs simultaneously with individual strength controls
• Two Modes:
  • Standard mode: Single strength value per LoRA, applied uniformly via ComfyUI's built-in loader
  • LTX mode: Per-layer strength control for LTX2 video/audio models (enable via the LTX checkbox)
• LTX Layer Controls (when LTX mode is enabled):
  • STR (Strength Model): Overall model strength
  • Vid (Video): Video attention layers
  • V2A (Video-to-Audio): Cross-modal attention (video to audio)
  • Aud (Audio): Audio-specific layers
  • A2V (Audio-to-Video): Cross-modal attention (audio to video)
  • Other: Remaining network components
• Intuitive UI: Drag-and-drop row reordering, toggle enable/disable, click-to-edit or drag-to-slide values
• Drag-and-Drop File Import: Drop LoRA files onto the node to quickly add them — matches by filename against your existing LoRA library
• Missing LoRA Detection: LoRAs that no longer exist on disk are shown with red strikethrough text on session load, so you can spot moved or deleted files at a glance
• rgthree LoRA Info Integration: Right-click any LoRA name to open the rgthree LoRA info dialog (requires rgthree-comfy). LoRAs with fetched info display a circled info badge next to their name (green for Civitai data, gray for local info). Fetching new info from Civitai in the dialog updates the badge immediately on close
• Optional Model/CLIP Input: Use as a standalone LoRA manager even without model or CLIP connected
• JSON Output: Export a rich JSON structure of all selected LoRAs for external processing
• Raw Config Editor: Click the cog button to copy/paste the entire LoRA configuration as JSON
• Sidecar Metadata Support: Automatically loads metadata from .json sidecar files alongside LoRA weights
• MultiLoRA Cycle (companion node): Automatically iterate through LoRAs at multiple strengths for batch testing — ideal for evaluating training checkpoints
• Parse JSON (companion node): Convert the lora_data JSON output into a Python object for use with scripting nodes like rgthree's Power Puter

Installation

1. Clone or download this repository into your ComfyUI custom_nodes directory:

   cd ComfyUI/custom_nodes
   git clone https://github.com/phazei/ComfyUI-MultiLoraLoader.git

2. Restart ComfyUI or reload the browser page.

3. The node will appear in the node menu under Loaders > Multi LoRA Loader. You can also search for "power lora loader", "lora stack", or "multi lora".

Usage

Basic Workflow

1. Add a Multi LoRA Loader node
2. Connect model and/or CLIP inputs (both optional)
3. Click + Add LoRA to add rows
4. Click the LoRA name field to open a menu of available LoRAs
5. Adjust strength values by:
   • Dragging left/right for fine control
   • Single clicking to type an exact value
6. Toggle the dot on/off to enable/disable individual LoRAs
7. Drag rows by the grip handle to reorder
8. Click the X to delete a row
9. Enable the LTX checkbox to show per-layer strength columns

UI Elements

Element	Action	Purpose
LTX (Checkbox)	Click	Toggle LTX per-layer mode on/off
Cog	Click	Open raw JSON editor for copy/paste/sharing
Grip	Drag vertically	Reorder LoRA rows
Dot (Toggle)	Click	Enable/disable the LoRA
LoRA Name	Left-click	Open LoRA selection menu
LoRA Name	Right-click	Show LoRA Info (requires rgthree-comfy)
Info badge	—	Indicates fetched LoRA info (green = Civitai, gray = local)
Strikethrough	—	LoRA file not found on disk (hover for tooltip)
STR / Vid / V2A / Aud / A2V / Other	Drag or click	Adjust strengths
X (Trash)	Click	Delete the row
+ Add LoRA	Click	Add a new empty row
Drop zone	Drop files	Drop LoRA files to add by filename match (files must already be in your loras directory)

Note on drag-and-drop: No files are copied — this is a shortcut for selecting LoRAs from the menu. Dropped files are matched by filename (case-insensitive) against LoRAs already in your ComfyUI models/loras directory. If a file isn't found, an alert lists the unmatched filenames.

Inputs and Outputs

Inputs (all optional):

• model: Model to apply LoRAs to
• clip: CLIP to apply LoRAs to (standard mode only)
• load_options: Connect a MultiLoRA Cycle node here for automated iteration

Outputs:

• model: The input model with all active LoRAs applied
• clip: The input CLIP with LoRAs applied (standard mode only; LTX mode passes CLIP through unchanged)
• lora_data: JSON string containing metadata for all selected LoRAs (enabled and disabled)

rgthree Integration

This node integrates with rgthree-comfy for LoRA metadata viewing:

• Right-click any LoRA name and select Show LoRA Info to open rgthree's info dialog with Civitai data, trained words, sample images, and editable notes
• LoRAs that have fetched info display a small circled info badge next to their name:
  • Green badge: Civitai data has been fetched
  • Gray badge: Local info file exists
  • No badge: No info available yet (use the dialog to fetch it)

Info is cached after first fetch, so toggling, reordering, or adding LoRAs won't cause badge flicker. Closing the info dialog after fetching new data from Civitai automatically refreshes the badge.

If rgthree-comfy is not installed, the info features are simply not shown.

Example Output

The lora_data output includes all LoRAs with selected weights:

[
  {
    "name": "my_lora.safetensors",
    "path": "/path/to/my_lora.safetensors",
    "enabled": true,
    "strength_model": 1.0,
    "video": 0.8,
    "video_to_audio": 1.0,
    "audio": 0.5,
    "audio_to_video": 1.0,
    "other": 1.0,
    "metadata": {...}
  }
]

Sharing and Batch Editing

Using the Config Editor (cog button)

The easiest way to share or batch-edit your LoRA configuration:

1. Click the cog button in the header of the node
2. Copy the compact JSON from the dialog
3. Edit it in your text editor
4. Paste the edited JSON back and press Enter
5. Invalid JSON is ignored and the config stays unchanged

Example config:

[{"on":true,"lora":"style_lora.safetensors","str":1.0,"vid":0.8,"v2a":1.0,"aud":0.5,"a2v":1.0,"other":1.0},{"on":false,"lora":"face_detail.safetensors","str":0.7,"vid":1.0,"v2a":1.0,"aud":1.0,"a2v":1.0,"other":1.0}]

Workflow JSON Editing

You can also hand-edit workflows in the saved JSON file:

1. Open the workflow JSON file
2. Locate your node's properties.lora_data field
3. Edit the LoRA list directly
4. Save and reload in ComfyUI

Companion Nodes

MultiLoRA Cycle

Automatically iterates through LoRAs in your Multi LoRA Loader at multiple strength values across queue runs. Designed for batch-testing LoRA training checkpoints — load all your checkpoints into the loader, set a list of strengths to test, and let it cycle through every combination.

Setup

1. Add LoRAs to your Multi LoRA Loader (via the UI, cog button, or drag-and-drop)
2. Add a MultiLoRA Cycle node (found under Loaders)
3. Connect its load_options output to the loader's load_options input
4. Set your test strengths as a comma-separated list (e.g. 0.5,0.75,1.0,1.25)
5. Set mode to increment
6. Enable auto-queue in ComfyUI and hit Queue

The cycle node walks through every LoRA at every strength, one per queue run. The loader's UI updates in real time to show which LoRA is active and at what strength.

Widgets

Widget	Type	Description
strengths	STRING	Comma-separated strength values to test (e.g. 0.5,0.75,1.0). Negative and zero values are valid.
lora_index	INT	Current LoRA position (1-based). Auto-incremented when all strengths for the current LoRA are exhausted.
strength_index	INT	Current strength position (1-based). Auto-incremented each queue run.
mode	COMBO	increment auto-advances each run; fixed stays on the current indices.
loop	BOOLEAN	When all combinations are exhausted, wrap back to the start instead of stopping.

Display

The node shows a status line at the bottom:

my_training-000800 (3/7) | STR: 0.75 (2/4)

This tells you: LoRA 3 of 7 is active, at strength value 2 of 4 (0.75). When all combinations are done and loop is off, it shows Done (7 LoRAs x 4 strengths).

How Cycling Works

• Each queue run increments strength_index
• When strength_index exceeds the number of strengths, it resets to 1 and lora_index advances
• When lora_index exceeds the number of LoRAs: wraps to 1 if loop is on, or stops (all LoRAs disabled) if off
• The first queue run after setting up uses the current indices as-is (no increment on the first run)

Stopping and Resuming

• Stop auto-queue at any time — the indices stay where they are
• Manually adjust lora_index or strength_index to jump to any position
• The display updates immediately when you change the indices

Saved Metadata

When output files are generated, the cycle node's mode is automatically saved as fixed in the workflow metadata. This means loading a workflow from an output file won't accidentally start auto-cycling — you'll see the exact LoRA and strength that produced that output.

---

Parse JSON

Converts a JSON string into a native Python object (list, dict, number, etc.). Connect it to the Multi LoRA Loader's lora_data output to feed structured LoRA data to nodes that accept any-type inputs.

Port	Type	Description
json_string (input)	STRING	A JSON-encoded string (e.g. the lora_data output)
data (output)	* (any)	The parsed Python object

Usage with Power Puter

This node pairs well with rgthree-comfy's Power Puter node, which lets you write inline Python expressions in your workflow. Connect the Parse JSON output to a Power Puter input to extract and format LoRA information.

Example: Generate a text overlay showing active LoRA names and strengths:

'\n'.join([l['name'].split('\\')[-1].rsplit('.', 1)[0] + ': ' + str(l['strength_model']) for l in a if l['enabled']]) if a else 'No active loras'

This produces output like:

my_training-000800: 0.75
style_lora: 1.0

---

Migration

This node was previously called Power LTX LoRA Loader Extra. If you load an old workflow:

1. The old node will appear with a "Node Renamed" notice and your LoRA data visible in a text box
2. Copy the JSON data from the text box
3. Add a new Multi LoRA Loader node
4. Click the cog button on the new node and paste the data in
5. Enable the LTX checkbox if your workflow used LTX layer-specific strengths
6. Delete the old deprecated node

Troubleshooting

LoRA name shows red strikethrough

• The LoRA file was not found in your models/loras directory
• It may have been moved, renamed, or deleted
• Hover over the name for a "LoRA file not found" tooltip
• The LoRA will be skipped during execution but its configuration is preserved

LoRA doesn't appear in the menu

• Ensure the LoRA file is in your ComfyUI models/loras directory
• Refresh the page or restart ComfyUI

Disabled LoRAs don't affect the model

• Disabled LoRAs (toggle off) are kept in your configuration but not applied
• Your full configuration including disabled LoRAs is preserved in the workflow

Can't shrink the node after expanding it

• Drag the node's resize handle to make it narrower
• The node enforces a minimum width of 500px

LoRA Info / right-click menu not working

• Requires rgthree-comfy to be installed
• If rgthree is not available, the info features are silently disabled

License

Apache 2.0. See LICENSE file.

Credits

Built on the Power Lora Loader concept by rgthree, extended with multi-LoRA management, LTX2 layer-specific strength control, and rgthree info dialog integration.

Contributing

Contributions, bug reports, and feature requests are welcome! Please open an issue or submit a pull request at github.com/phazei/ComfyUI-MultiLoraLoader.