ComfyUI Refresh LoRA List: A Detailed Guide to LoRA Management in ComfyUI

UPDATE! Access Tensorboard is a very simple node that launches a URL to see data about the logs created during training. Click the node once (after, during, or even before training!), then copy-paste the URL that it wrote in the command prompt. The Advanced node is full of many recommended features. Some are still missing, but they are here in the code, you just have to change their values manually! Not everything was tested, I personally don't even know what most of these do. This node is now confirmed to work with LCMs, SD2.0 and SD Turbo models.

DISCLAIMER: This is my first "big" custom node. More specifically, it's the first I publish that has a lot of requirements. Therefore, it could conflict with other custom nodes. This custom node lets you train LoRA directly in ComfyUI! By default, it saves directly in your ComfyUI lora folder. I'll link my tutorial. In short, you have to put it in custom_nodes and install the requirements (from requirements_win.txt if you're on Windows). Copy the path of the folder ABOVE the one containing images and paste it in data_path. Finally, just choose a name for the LoRA, and change the other values if you want. That way you just have to gather images, then you can do the captioning AND training, all inside Comfy!

But I have no idea about SDXL. If someone could test it and confirm or infirm, I’d appreciate ^^. The very first version I published online had very strict requirements, which made conflicts likely. I've reduced requirements to the bare minimum, but conflicts can still arise.

LoRA Fundamentals and Why They Matter

"No module X found" : This error happens when Python can't find the required modules for the running program. But it can also happen if you installed the requirements in the wrong folder! "Something about cuda.dll missing" : there's a lot of variants of this error, but they're all the same problem (hopefully): it happens if you haven't installed CUDA properly (or, again, if it's not installed in the right folder). I own a Windows 10 machine with a RTX 3060. Even the Advanced node doens't include all inputs available for LoRA training, but you can find them all in the script train.py!

This is based off an existing project, lora-scripts, available on github. I took that project, got rid of the UI, translated this “launcher script” into Python, and adapted it to ComfyUI. For updates, check this post: linkLatest video:[Update 0.5.7] Performance Boost & Search Feature🚀 Major Performance Improvements:Implemented multi-layer caching and cache preloadingAdded file system monitoring with incremental updatesIntroduced pagination API with infinite scroll support🔍 Search Functionality: New search feature to quickly find LoRA models🐛 Bug Fixes: Various stability and performance improvements[Update 0.5.6] New Features and Optimizations🛠️ Code Refactor: The codebase has been restructured to improve readability and maintainability, making it easier to manage and extend in future updates.🚀 Frontend Enhancements: Significant performance improvements and refined user experience, including a more intuitive process for selecting and copying trigger words.🔘 New Menu Button: A button has been added to the ComfyUI menu. Clicking it will open the LoRA Manager interface in a new window for quicker access.🛠️ Extra LoRA Paths - Support for additional LoRA paths via extra_model_paths.yaml is now available, offering greater flexibility in managing LoRAs located outside the default directory.[Update 0.5.3] Improved Preview Handling & Trigger Words Support✅ Smarter Preview Image Handling - The manager now scans for existing local preview images and uses them automatically. If a local preview is found, it won’t re-download one from CivitAI when fetching model details.📝 Trigger Words in LoRA Details - Now, trigger words are displayed in the LoRA details window, making it easier to see and copy it.

Hey fellow ComfyUI users! Do any of you struggle with managing a growing collection of LoRA models? I realized I kept forgetting which LoRAs I had downloaded and what each one actually did when building workflows. If this sounds familiar, I've got something to share! Over the weekend, I built ComfyUI Lora Manager - a simple solution to visualize and organize your local LoRA models. What features would you want next?

Core LoRA Concepts and Quick Start

LoRA stands for Low Rank Adaptation. Quick answer: LoRA is a small add-on model that attaches to your base model in ComfyUI. It steers how the model generates images - teaching it a new style, character, or fix - without retraining the whole model. LoRA files are small (typically 5-150 MB) compared to full models (4-20 GB). They download fast, swap instantly, and combine easily. This is why they have become one of the most-used tools in the ComfyUI ecosystem.

Why Use LoRA in ComfyUI - The Benefits

  • Consistent style: Lock in an art style - pencil sketch, watercolor, anime - and reproduce it consistently without rewriting your prompt every time.
  • Consistent character: A character LoRA remembers a specific person's face, outfit, and proportions so every generation matches.
  • Fix common problems: Dedicated LoRAs improve hands, faces, anatomy, or lighting - areas where base models frequently struggle.
  • Faster generation: Turbo/lightning LoRAs produce quality results in 4-10 steps instead of 30-40, cutting generation time significantly.

Using LoRA in ComfyUI involves three steps: download the file from a trusted source, place it in the correct folder, and load it with a LoRA node in your workflow.💡 Tip: Download the example workflow first. All LoRAs shown in this guide are referenced in the workflow file. Download Basic LoRA Workflow

Step 1 - Download from a trusted source

Only use reliable platforms: Civitai, Hugging Face, ModelScope. Accepted formats: .safetensors, .pt, .pth. Avoid unknown formats.

Step 2 - Place the file in the correct folder

Place in ComfyUI/models/loras. After placing the file, refresh the LoRA list so ComfyUI can see it.

Step 3 - Add a Load LoRA node to your workflow

Right-click the canvas and search for Load LoRA. Connect it between your checkpoint loader and your KSampler - the LoRA node takes MODEL and CLIP as input, and outputs modified versions of both.

The Load LoRA node sits between the checkpoint loader and the KSampler. Step 4 - Verify the LoRA is working: Run a test prompt and compare output with and without the LoRA active. If you see no difference, check that the LoRA is compatible with your base model.

Applying LoRA in Your Workflows

There are two main approaches: a single LoRA for one focused change, or multiple stacked LoRAs for combined effects.

Single LoRA - the simplest approach: Start with a weight of 0.5 and adjust from there. Use model strength and clip strength, often setting them to the same value. For FLUX-based LoRAs, clip strength often has little effect.

Multiple LoRAs - LoRA stacking: Chain multiple Load LoRA nodes together. This lets you combine effects - for example, a photography style LoRA plus a hand fix LoRA. When stacking, lower each individual weight (0.3-0.5 each) so they blend naturally. One LoRA at full strength will overpower everything else.

Managing LoRAs with Extensions

ComfyUI-Lora-Manager is a free extension that replaces the plain filename dropdown with a visual library - previews, trigger words, and saved combinations included. It provides a visual grid of every LoRA and checkpoint with a preview thumbnail, a Trigger Word Toggle, automatic previews from CivitAI via API key, and LoRA Recipes that save combinations. Install it through ComfyUI Manager and use it from the menu bar to open the visual library in a new tab.

Tip: You don’t need Lora Manager to follow the rest of this guide - the standard Load LoRA node works everywhere. It’s worth installing once your LoRA collection grows past what a dropdown can handle comfortably.

Best Practices: Settings, Weights, and Troubleshooting

The weight (strength) slider is the most important setting on a LoRA node. It controls how strongly the LoRA influences output - from a subtle nudge to a complete style takeover. Weight reference: 0.1-0.3 subtle, 0.4-0.6 balanced, 0.7-0.9 strong, 1.0+ overpowering. When stacking, lower each individual weight to blend.

Model strength vs clip strength: Set them to similar values in most cases. If the visual style should appear without affecting prompt interpretation, you can lower clip strength independently. For FLUX LoRAs, clip strength often has little effect because FLUX conditions primarily through its T5 text encoder rather than CLIP.

Common causes if LoRA has no effect or errors: model mismatch, wrong file location or format, failure to refresh after adding a new file, too high weight, or misconnected nodes. Solutions: verify the base model compatibility, ensure the file is in ComfyUI/models/loras and in a supported format, refresh the LoRA list, adjust weight, and confirm proper node connections.

Frequently Asked Questions: LoRA is a small model file that attaches to your base model and changes how it generates images - adding a style, character, or improvement. Place LoRA files in ComfyUI/models/loras, use supported formats, and refresh the list after adding new files. For stacking, start with 0.5 and adjust in 0.1 steps. ComfyUI-Lora-Manager is a helpful companion but not mandatory, especially for moderate collections.

Practical Workflow Tips

1) Start with a single LoRA and a moderate weight (0.5). 2) If needed, stack a second LoRA with adjusted weights (0.3-0.5). 3) Regularly refresh the LoRA list after adding new files. 4) Consider using LoRA Manager for large libraries to keep track of trigger words and recipes. 5) Test prompts with and without LoRA to verify compatibility and effect.

  1. Place LoRA files in the correct folder.
  2. Refresh the LoRA list to make them visible.
  3. Load LoRA via the Load LoRA node and connect to MODEL and CLIP inputs.
  4. Experiment with weights and stacking to achieve the desired look.
ComfyUI LoRA library visual overview

Tables and Quick Reference

AspectGuidance
LoRA sizeTypically 5-150 MB
File formats.safetensors, .pt, .pth
Recommended starting weight0.5
Stacking weights (per LoRA)0.3-0.5 each for blending
Refresh methodsIcon refresh, hard refresh, restart server

tags: #comfyui #refresh #lora #list