Guide to Using Model Trainer

What does the Model Trainer do?

Unlike traditional model training or fine-tuning, our Model Trainer uses the DreamBooth process to add a custom "instance prompt" token to a base model (FLUX.1 [dev]). 

This token is unique to your custom model and, when enabled, automatically invokes your Character or Style when creating new images with AI Generator. You don’t need to manually add the token to your prompts.

Training images

AI images

How to Use Model Trainer?

Creating your own AI model is easier than you might think; it only takes a few clicks. Here’s how to do it.

Step 1. Select Your Model Type

Head to the Model Trainer through the tool picker or the AI Generator, or click here. 

First, choose what type of model you want to train: Style or Character.

• If you want to create a model that reflects a particular aesthetic (such as a painting or photography technique), select Style;

• If you want to create a custom model based on a specific person (e.g., yourself), choose Character.

Next, give your model a name. This is just to help you manage your models more easily; the model's name doesn't affect the training process at all.

Step 2. Choose Gender (for Characters)

If you go with a Character model, you’ll need to pick a gender: Male, Female, or Neutral.

Step 3. Upload Training Photos

Now, the most important part: uploading your photos. This is where the quality of your model takes shape, so follow these guidelines for the best results:

• Upload a minimum of 3 photos. Aim for 10-20 for the best results, but fewer may work—though it could affect model quality. The number of training images will affect training time.

• Each image should have a resolution of at least 1 Mpx (~1024x1204 pixels or more). You’re not limited to any specific aspect ratio: Model Trainer employs multi-resolution training, meaning images will be cropped to the closest FLUX.1 [dev]-supported resolution if needed.

Make sure your selection is diverse:

• For a Character, upload a range of photos with different angles, expressions, and poses. Make sure to include different backgrounds, lighting, close-ups, and full-body shots. Stick to solo photos of the subject. No other people, sunglasses, hats, or excessive makeup.

• When training a Style, include a diverse set of images of different objects, people, and scenes made in your target aesthetic.

It’s important to follow these rules when uploading your images:

• No harmful or offensive content. You’re not allowed to create anything inappropriate, including deepfakes.

• Only upload images you own or have full legal rights to use. 

• If you’re training a model based on a real person, get their permission to use their images for this purpose.

Step 4. Start Training

After uploading your images, you’ll see an estimate of how long the training will take. Double-check your choices and hit “Train model.” Once you agree to the aforementioned rules, we'll begin the training process.

Managing Your Character and Style Models

Changing Model Names

All you need to do to rename a model is:

1. In the Model Trainer, click on the three dots next to the model you'd like to rename.

2. Choose "Update Name" from the dropdown menu.

3. Input the new name and press "Update model".

Deleting Models

To delete a model:

1. Open the Model Trainer and click on the three dots next to the model you want to get rid of.

2. Press "Delete" and confirm your choice. Keep in mind that this operation is completely irreversible. 

After deletion, you will permanently lose the ability to generate images with this Character or Style. However, any existing images created with it will still be accessible.

That’s everything you need to know to train and manage your AI models. We’re excited to see what you create!