What Is Style Training?
Style training allows you to create your own custom AI style.
It works by fine-tuning an existing AI model using a smaller set of images (your dataset). A dataset is simply a collection of images that represent the look, aesthetic, or subject you want the AI to learn.
Once trained, the AI can generate new images that match the style or characteristics of your dataset.
Train Your First Style
1️⃣ Gather Your Training Data
Collect 5–60 images of a consistent style, object, or aesthetic
The more high-quality images you include, the better your results
Crop images to 1:1 (square) to avoid distortion after training
Make sure your images follow our Content Policy
Avoid mixing very different styles in one dataset
A strong, consistent dataset leads to stronger results.
2️⃣ Upload Your Dataset
Go to the Styles tab
Select My Styles
Tap Create Style
Upload your images
You can crop your images directly inside starryai using the built-in crop tool.
After uploading:
Give your style a name
Select the most relevant category
Confirm to begin training
Training a style costs 50 Lumens.
Here’s an example of an image that was created using this style
Prompt: Portrait of a handsome man
3️⃣ Start the Training Process
Training typically takes around 45 minutes
During busy periods, it may take up to 2 hours
If training fails after 2 hours, your Lumens are automatically refunded
Once training is complete, you’ll receive a notification.
After that, you can:
Use your style to generate new images
Publish your style to share it with other users
Example
Prompt: Portrait of a handsome man
(Using a trained custom style)
The output will reflect the aesthetic and characteristics learned from your dataset.