- Flux is a new open-source model to generate AI images, and you can train the model on your own images using the LoRA technique.
- Fal.ai lets you train a Flux LoRA model in a straightforward manner. You don't need a cluster of GPUs or technical knowledge to train a custom model.
- In my testing, the Flux LoRA model performed exceptionally well, and we have included some AI-generated image samples here.
While you can generate cool AI images using Flux for free, training the model using the LoRA (Low-Rank Adaptation) technique allows you to fine-tune the model on your specific dataset. Basically, if you use Flux LoRA training, you can generate AI images in a particular style or train the model on your own images. Thankfully, there is now a straightforward way to train Flux LoRA without needing a beefy GPU or technical knowledge. Here’s how to get started and how it works.
Notable Points You Should Know
- First, you need to sign up for fal.ai, for which you need a GitHub account. Go ahead and create an account for free here.
- You need to top up at least $10 on fal.ai. For one training run (1000 steps), it costs $5. You can add balance to your fal.ai account using the link here.
- It takes about 20 minutes to train a Flux LoRA model on fal.ai.
- Once the training is complete, you can generate up to 29 images for $1.
How to Train Flux LoRA on Your Images
- Go ahead and open fal.ai’s Flux LoRA training page and click the “Sign in to run.” Log in with your GitHub account.
- Next, I am assuming you have already added a $10 balance to your account.
- Now, go ahead and add 6-12 images on which you want to train Flux LoRA. You can add more images to personalize the model even further.
- Now, expand “Additional Settings” and add a trigger word. During prompting, you can use this keyword to refer to the character or an artistic style. Since I am training the model on my own images, I have given my name as the trigger word. Keep everything else as default.
- Finally, click on “Start” at the bottom to start the training process, and it will last for about 20 minutes.
- Once the training process is complete, click on “Run inference.”
- Now, you can start entering text prompts to generate images. Add the trigger word to generate images of your own or refer to the artistic style. Here are some of my samples:
In my testing, I found the Flux LoRA training process quite user-friendly. If you are a developer, you can use the LoRA checkpoint and integrate it into your services.
Since the base model is FLUX.1 [dev], it’s really good at text rendering, as we mentioned in our Flux vs Midjourney comparison. Not to mention, it handles hands and fingers pretty well. Anyway, that’s all from us. If you have any questions, let us know in the comments below.
