How to Use Flux LoRA's in ComfyUI: A Complete Walkthrough

In this tutorial, you’ll learn how to harness the full potential of Flux LoRA within ComfyUI — a powerful, node-based interface for AI image and video generation. Flux is a cutting-edge diffusion model known for producing highly realistic, prompt-accurate results. But when you need more stylized, artistic, or character-specific outcomes, that’s where LoRA (Low-Rank Adaptation) models shine. This guide will show you how to combine the Flux Dev base model with one or more LoRAs to enhance image generation, explore new styles, and gain fine-grained control over your outputs. We'll dive into loading LoRAs, stacking multiple models, adjusting strengths, previewing outputs, and even where to find the best LoRAs online. Whether you're building your first AI workflow or optimizing complex pipelines, this tutorial will give you practical, hands-on guidance to get the most out of Flux LoRA in ComfyUI.

LoRA, or Low-Rank Adaptation, is a technique used to fine-tune large diffusion models like Flux Dev without retraining them from scratch. Rather than modifying the entire base model, a LoRA acts like a lightweight “style patch” — applying targeted changes that shift the style, content, or structure of the generated image.

Why use LoRAs with Flux Dev?

Because while Flux excels at photorealistic, prompt-accurate generation, it isn’t always the most imaginative model out of the box. LoRAs fill that creative gap by letting you:

• Stylize your images with a particular art direction

• Recreate consistent characters or facial features

• Change clothing, body types, or poses

• Inject specific environments or themes into a scene

LoRAs don’t replace your base model — they enhance it. You can fine-tune how much each LoRA influences the output, and because they’re modular, you can experiment freely without needing to bake changes into the model itself.

To better understand this, let’s look at a clear example comparing an image generated by the base Flux Dev model alone versus the same prompt enhanced with an anime-style LoRA applied.

Base Prompt: A hyperrealistic, busty goddess stands confidently against a glowing twilight sky, captured from waist up in an extreme close-up. Her sheer lace top reveals her voluptuous cleavage and toned midsection. Long silver hair cascades down her back, sparkling under cinematic soft lighting. Her crystal-blue eyes gaze directly at the viewer, radiating warmth and seduction. Ethereal floating lights and subtle mist add a dreamlike, magical aura. Photorealistic skin texture, ultra-detailed face.

This side-by-side comparison highlights exactly how a LoRA can transform and stylize the original output while still building on the strengths of the base model.

Before we dive into generating images with Flux Dev and LoRAs inside ComfyUI, you’ll need to set up your environment with the correct model version and files. For this guide, we’ll start with the Flux Dev FP8 version — our recommended starting point for its excellent image quality, high performance, and straightforward setup inside ComfyUI. (If you're working with limited VRAM or need a lighter setup, we’ll cover the GGUF version later, which is optimized for lower VRAM environments.)

1. Install ComfyUI

Flux Dev runs inside ComfyUI, a flexible and powerful node-based interface for AI image generation.\

If you haven't installed it yet, choose the setup that fits your system:

• Local Installation
  Ideal for users with a PC or workstation (12GB+ VRAM recommended).
  👉 Follow our Local ComfyUI Setup Guide.

• RunPod Deployment
  Perfect for cloud-based GPU access with minimal local setup.
  👉 How to run ComfyUI on RunPod.

2. Download the Required Files

To run Flux Dev FP8 in ComfyUI, you’ll need to manually download and place the following files. These are not bundled with ComfyUI by default.

Flux Dev FP8 Model Checkpoint

This is your main model file for Flux Dev:

• 🔗 flux_dev_fp8.safetensors

• 📁 Save to: ComfyUI/models/checkpoints/

3. Verify Your Folder Structure

After downloading and placing the files, your directory should look like this:

ts
1📂 ComfyUI/
2├── 📂 models/
3│   └── 📂 checkpoints/
4│       └── flux_dev_fp8.safetensors

Before we dive into using LoRAs and customizing your generations, the next step is to ensure your workflow is fully functional by fixing any missing nodes or dependencies. Once that’s done, we’ll explore how to find and apply LoRAs to bring more creativity and control to your Flux Dev outputs.

Now that your environment and model files are in place, it’s time to load and configure the Flux Dev FP8 LoRA workflow in ComfyUI. This setup ensures that all components work seamlessly together for optimal image quality, smooth LoRA stacking, and stable performance. Once configured, you'll be ready to start experimenting with Flux Dev and LoRAs using real prompts.

1. Load the Flux Dev LoRA Workflow

Start by loading the pre-built Flux Dev FP8 LoRA workflow:

👉 Download the provided Flux Dev FP8 Lora Workflow JSON file workflow file and drag it into your ComfyUI canvas.

This workflow template includes all required nodes for Flux Dev with LoRA support, pre-arranged for easy customization and immediate testing.

2. Installing Missing Nodes for Flux Dev LoRA Workflow

When you load the Flux Dev LoRA workflow, you may see red nodes — this means some custom nodes are missing. In this case, we need to install the Power Lora Loader (rgthree) node.

Follow these quick steps:

• Click the Manager button (top right).

• Click Install missing custom nodes — ComfyUI will scan for any missing nodes.

• In the list, search for rgthree-comfy and click Install.

  

• Click Restart (bottom left), confirm the restart, and wait for reconnection.

• Once reconnected, refresh your browser tab.

That’s it — your workflow is now fully ready for LoRA stacking.

Where Do Flux LoRAs Get Loaded?

With the Power Lora Loader node now installed, you’ll see a dedicated node inside your workflow where LoRAs can be added.

This node allows you to:

• Load multiple LoRAs at once

• Adjust individual strengths

• Stack LoRAs together for more advanced combinations

This is where you’ll load and manage your LoRAs directly inside ComfyUI.

To start using LoRAs with Flux Dev in ComfyUI, your first step is discovering the right LoRA models. The best source for this is Civitai and depending on how you browse, there are two different filter systems to be aware of.

Option 1: Using the Main Models Page

If you go directly to the Models page on Civitai, click the filter icon in the top-right corner. Then set the following options:

• Time Period: All Time

• Model Types: LoRA

• Base Model: Flux.1 D

This will give you a curated list of all the LoRAs specifically trained for Flux Dev — no guesswork required.

Option 2: Using the Search Bar (with Sidebar Filters)

If you use the search bar instead — for example, typing in "fantasy" or "anime" — make sure to refine your results using the sidebar filters that appear on the left:

• Base Model: Select Flux.1 Dev

• Model Type: Select LoRA

⚠️ Without these filters, your results might show LoRAs trained on completely different base models (like SDXL), which may not work properly with Flux Dev in ComfyUI.

Once you’ve filtered your results and found a promising LoRA, it’s time to download it and see it in action.

Now that you know how to search for Flux-compatible LoRAs on Civitai, let’s walk through the full process of installing and preparing one for use inside ComfyUI with Flux Dev.

For this example, we’ll use the Cyberpunk Anime Style LoRA, which gives your generations a vibrant neon sci-fi or anime look.

Step 1: Open the LoRA’s Page

Once you’ve found a LoRA you want to try — like Cyberpunk Anime Style — click into its individual model page on Civitai. This page contains everything you need to properly install and use the LoRA.

Step 2: Check for Trigger Words

Almost every LoRA comes with one or more trigger words — special terms you need to include in your prompt to activate the LoRA’s style effect. If you forget these, the LoRA may technically load, but your image will simply fall back to Flux Dev’s native photorealism.

You can usually find the trigger words:

• In the model details section

• Embedded inside example prompts

• In comments or notes from the creator or other users

💡 Example for Cyberpunk Anime Style:\

This LoRA uses the trigger words:

• anime

• cyberpunk

You can include either word individually or both together for a hybrid effect. If you skip these trigger words, the LoRA will have no visible impact — even though it technically loaded into ComfyUI.

Step 3: Download the LoRA and Place It in Your LoRA Folder

Click the Download button on the LoRA’s Civitai page to get the .safetensors file.

Before moving it into ComfyUI, you can rename the file to anything you like — just make sure it’s a unique, simple name you can easily remember and identify later. For this example, we’ll name the file: CyberpunkAnime.safetensors

ts
1📂 ComfyUI/
2└── 📂 models/
3    └── 📂 loras/
4        └── CyberpunkAnime.safetensors

This step ensures the LoRA will be recognized by ComfyUI the next time it starts or reloads.

With your Cyberpunk Anime Style LoRA safely placed into the correct folder, it’s now available inside ComfyUI — but before it can affect your generations, you still need to:

1. Load the LoRA inside the Power Lora Loader node

2. Activate it with the correct trigger word inside your prompt

This gives you full control — you decide exactly which LoRAs to activate for each generation, and how strongly each one should influence the output.

Now that your workflow is fully prepared, it’s time to load your LoRA, enter the trigger words, and run your first generation. This simple process connects everything you’ve set up so far and lets you see the LoRA’s effect directly in your output.

Step 1 — Load the LoRA in the Power Lora Loader Node

Inside your loaded workflow, locate the Power Lora Loader (rgthree) node.

• Click on the node.

• In the dropdown list, select your downloaded LoRA — in this case:
  CyberpunkAnime.safetensors

• Leave the strength at 1.0 for now (you can adjust it later to fine-tune the effect).

Step 2 — Enter the Trigger Word in the Positive Prompt

Simply selecting the LoRA doesn’t apply its effect automatically — you must still include the correct trigger word in your positive prompt. For this LoRA, you can use either anime or cyberpunk as trigger words, depending on which style you want to apply to your generation.

Examples — Anime Style Activation

Let’s start by testing the anime trigger word from the Cyberpunk Anime Style LoRA.

Anime Example 1

Prompt: anime, busty woman with fiery red hair styled in loose curls, wearing a low-cut white blouse partially unbuttoned, sitting against a backdrop of swirling smoke and floating neon butterflies. The lighting is warm, almost candlelit, enhancing her soft skin and inviting gaze.

Anime Example 2

Prompt: anime, a confident woman with long dark hair in a high ponytail stands mid-action in a cyber dojo, her exposed cleavage framed by a red and black futuristic martial arts outfit. Thick smoke curls from glowing incense burners and neon signs flicker in the background. Her eyes blaze with intensity as she looks at the viewer.

Anime Example 3

Prompt: anime, hyperrealistic busty magical nurse stands in a glowing clinic, enormous breasts pushing out of a tight, shiny white mini-dress with an open zipper exposing cleavage and underboob. Her waist is tightly cinched, emphasizing her exaggerated curves. Twin blonde ponytails swing over her shoulders. Her wide blue eyes stare directly at the viewer with a seductive smile. Magical glowing hearts float softly around her in a dreamy pastel-lit background. Skin texture, tight fabric, and light reflections rendered in extreme detail.

Examples — Cyberpunk Style Activation

Let’s now try some cyberpunk style by testing the cyberpunk trigger word from the Cyberpunk Anime Style LoRA.

Cyberpunk Example 1

Prompt: Cyberpunk, hyperrealistic ultra-busty woman floats in a holographic chamber, nude but covered strategically by glowing data ribbons wrapping tightly around her massive chest and wide hips. Her long platinum hair floats freely in zero gravity. Her glowing pink eyes stare directly at the viewer with a soft, seductive expression. Holographic energy pulses around her, with hyper-detailed skin pores, soft glows, and floating particles creating an extremely intimate, sensual sci-fi atmosphere.

Cyberpunk Example 2

Prompt: Cyberpunk, hyperrealistic busty woman lounges inside a luxury floating capsule, wrapped loosely in a shimmering nanofiber robe slipping off her shoulders, revealing her large bare breasts partially concealed by glowing circuit tattoos. Her silver hair is styled into sleek twin ponytails, and her teal eyes focus on the viewer with a gentle smirk. Surrounding her, glowing data streams and holographic UI panels create an intimate, high-end sci-fi atmosphere.

Cyberpunk Example 3

Prompt: Cyberpunk, highly-detailed biomechanical octopus with chrome-plated tentacles and embedded neon tubing. It floats weightlessly inside a massive zero-gravity research chamber surrounded by translucent data panels.

⚡ Note: All examples above rendered at 1280x720, 30 steps, ~20 seconds on RTX 4090.
🚀 No powerful GPU? Run Flux Dev on Cloud GPUs 👈

With your workflow ready, let’s take a closer look at the Power LoRA Loader node itself — the core of how we stack and manage LoRAs inside ComfyUI for Flux Dev.

1. Adding and Adjusting LoRA Strength

Inside the Power LoRA Loader node, you'll see an input field next to each loaded LoRA labeled Strength:

• The strength value typically ranges from 0.0 to 1.0.

• 0.0 means the LoRA has no effect (essentially disabled).

• 1.0 means the LoRA is fully applied at its maximum trained strength.

• You can enter any value between these to fine-tune the influence.

• When stacking multiple LoRAs, it's generally recommended to lower each strength slightly (e.g. 0.5, 0.6, 0.7), as applying all at full strength can sometimes lead to overbaked or conflicting results.

• This allows you to blend styles and influences more smoothly while maintaining stability and coherence in your generations.

2. Viewing LoRA Details and Trigger Words

A very useful feature:

• Right-click on any loaded LoRA inside the node and select Show Info.

• A popup window will appear with detailed information about the LoRA.

• In this window, you’ll find a Fetch From Civitai button.

  • Click this, and the node will automatically pull metadata directly from Civitai.

  • You’ll see where the LoRA was trained, its base model, example images, and most importantly: the recommended trigger words / trained words.

    

• If you’re unsure which trigger words to use, simply select the word from the list, then click Copy next to the suggested tags (found under the trained words on the left). You can then paste them directly into your positive prompt.

What’s Next: Exploring Multiple LoRAs

Now that you’ve mastered loading a single LoRA, it’s time to take things further. Next, we’ll explore stacking multiple LoRAs — focusing on realistic ones that enhance faces, modify body features like breasts, and add finer details. This approach helps you achieve the best, most believable results with Flux Dev.

Now that you’ve learned how to load and apply a single LoRA, it’s time to explore how combining multiple LoRAs can give you far greater creative control and realism. By stacking LoRAs, you can tweak different aspects of your image—from body shape to stylistic effects—to craft exactly the look you want.

For this example, we’ll use three complementary LoRAs:

1. Flux Skinny Petite
   🔗 [Civitai Link]

   • Trigger Word: skinny

   • Slims down the body shape to a petite, slender frame.

2. SamsungCam UltraReal
   🔗 [Civitai Link]

   • Trigger Word: s2ms8ng

   • Adds a natural, ultra-realistic phone-camera aesthetic, enhancing lighting and texture.

3. Flux Bolt-Ons (High Profile Breast Implants)
   🔗 [Civitai Link]

   • Trigger Word: b0lt0ns

   • Creates the look of lifted, enhanced breasts, adding sensual volume and shape.

Setting Up Multiple LoRAs in ComfyUI

To stack multiple LoRAs, simply select all three models—Flux Skinny Petite, SamsungCam UltraReal, and Flux Bolt-Ons—within your Power LoRA node. Adjust their strength values in the input fields, usually between 0 and 1. For the best, most natural results, avoid setting all strengths to 1, as this can cause visual artifacts. Instead, slightly lower each value—around 0.6 to 0.8—when stacking. Don’t forget to include all three corresponding trigger words (skinny, s2ms8ng, and b0lt0ns) at the start of your positive prompt to activate each LoRA properly. See the image below for my recommended node setup and strength settings if you want to follow along.

Now that we’ve loaded the Power LoRA Loader node, entered our carefully crafted prompt, and adjusted the strength values, it’s time to see the results in action. Below are some examples showcasing the impact of stacking the three LoRAs — Flux Skinny Petite, SamsungCam UltraReal, and Flux Bolt-Ons — with trigger words clearly placed at the start of the prompt.

With these examples, you’ve seen firsthand how stacking multiple LoRAs can elevate your images beyond what a single LoRA can achieve. By combining body shape modifiers, stylistic enhancements, and specific feature augmentations, you gain fine control over the aesthetics and realism of your creations.

If you have limited VRAM, Flux Dev offers GGUF model versions optimized for lower-resource setups. These versions come in different quantization levels (QX), where higher Q values require more VRAM. As a rule of thumb, the GGUF model file size (in GB) roughly corresponds to the VRAM needed to run it.

You can browse and download the right GGUF model version for your system from the official Flux Dev repository on Huggingface, selecting a balance between performance and quality based on your hardware.

Required Files and Their Locations

Example Folder Structure

ts
1
2ComfyUI/
3├── 📁 models/
4│   ├── 📁 diffusion_models/
5│   │   └── flux1-dev-QX.gguf
6│   ├── 📁 clip/
7│   │   ├── t5-v1_1-xxl-encoder-QX.gguf
8│   │   └── clip_l.safetensors
9│   └── 📁 vae/
10│       └── ae.safetensors

Workflow Setup

👉 Download the provided Flux Dev FP8 LoRA Workflow JSON file and drag it into your ComfyUI canvas.

This workflow template includes all required nodes for Flux Dev with LoRA support, pre-arranged for easy customization and immediate testing.

In conclusion, using Flux LoRA within ComfyUI unlocks a powerful and flexible approach to creating breathtaking AI-generated images. This tutorial has guided you through updating ComfyUI, loading workflows, downloading essential Flux and LoRA models, and effectively generating stunning visuals. The true magic comes with experimentation—don’t hesitate to explore different prompt styles, model combinations, and strength settings to find what best fits your unique creative vision. As you grow more comfortable with ComfyUI and Flux LoRA’s capabilities, your skill to craft distinctive, captivating images will expand dramatically. Whether you’re an artist, designer, or AI art enthusiast, the knowledge gained here will be a valuable asset on your creative journey. Happy creating!