Flux LoRA using FluxGym Tutorial: AI Image Training

Introduction

In this article, we'll walk you through the process of training Flux LoRAs using an easy-to-use application called FluxGym. I recently tested this app with an RTX 3060 having 12 GB of VRAM, and, while it worked, the training process was quite slow. It took over 10 hours to train with just 46 images. If you have 8 GB of VRAM, it might still be feasible, but you should anticipate a significant time investment.

Installation of FluxGym

To ensure a smooth experience, we will be using the one-click installation method. However, if you don't have Pinocchio installed, you can easily get it from Pinocchio.io. Here’s how to install FluxGym:

Access the Pinocchio app: Go to the Pinocchio app home.
Discover FluxGym: Click on "Discovery" and then select "FluxGY."
Download the App: Hit the download button; select 'Save as default,' and then click 'Download' again.
Install: Wait for a few seconds and then click 'Install.' Note that this process can take a while, depending on your internet speed, as it will download all models separately to the FluxGym folder.

After approximately 20 minutes, the installation should be completed, and you will be greeted with the FluxGym UI.

Training the Model

Now that FluxGym is set up, it's time to train a LoRA using 46 real images of a female model. Here’s what you need to do:

Upload Images: Start by uploading all the images you'd like to use for training.
Name Your Model: For this tutorial, we'll name our model "Alexa Grace."
Set a Trigger Word: We'll use "AG2655" as the trigger word for our model.
Configure VRAM: Set the VRAM to 12 GB.
Adjust Training Settings: To speed up the process, we can reduce the number of training iterations for each image to five. All other settings can remain default.
Add Captions: Scroll down and click on "Add AI Captions" with Florence, which will download the captioning model and automatically add captions to your images.

Once captioning is complete, click on “Start Training” to begin the process. You can monitor GPU performance to make sure training has commenced. Please bear in mind that the training might take several hours to finishing.

Completion and Testing

After about 10 hours, the training process will be completed. Navigate to the output folder to find your newly trained LoRA models. You should see four newly generated models there.

To test the trained models, exit the FluxGym and go to the Comfy UI. Using the Flux one Dev-based model, remember to include the trigger word in your prompts to achieve optimal results.

As highlighted, the images generated using the trained models should be almost 100% accurate relative to your training images.

If you’re interested in trying out the "Alexa Grace" model created in this tutorial, you can download it from Civit AI. Links can be found in the description below.

This concludes our tutorial on using FluxGym for AI image training. We hope you found this guide helpful. If you did, please consider subscribing and liking our channel, as your support truly makes a difference. Have fun creating!

Keywords

FluxGym
Flux LoRA
AI image training
RTX 3060
VRAM
Model training
Image captioning
Civit AI

FAQ

What is FluxGym?
FluxGym is an application designed to simplify the process of training Flux LoRAs using a user-friendly interface.

How long does it take to train a model using FluxGym?
The training duration can vary. In the tests conducted, it took about 10 hours to train using just 46 images with an RTX 3060 having 12 GB of VRAM.

Do I need a specific amount of VRAM to use FluxGym?
While 12 GB of VRAM is recommended for optimal performance, users with 8 GB may still manage, but they should expect longer training times.

Can I download the trained models?
Yes, trained models can be downloaded, and in this tutorial, the "Alexa Grace" model is available on Civit AI.

Is there a process for adding captions to training images?
Yes, you can use the built-in "Add AI Captions" feature with Florence for automatic captioning of images during the setup.