Training a background style in Layer allows you to generate consistent environments, rooms, or game scenes with a specific visual style. Whether you’re building matte paintings or isometric rooms, this guide covers how to structure your training data for the best results.
Start with Your Assets
Before starting, prepare your training assets. Use high-quality background images that share a consistent perspective, layout, and art direction. If you plan to generate empty rooms or upgrade variations, make sure your dataset reflects that goal - for example, use empty or minimally decorated rooms if that’s what you want to generate.
We recommend using around 25 high-quality images for best results. More is fine, but try to keep the visual style, angle, and level of detail consistent across your dataset to help the model learn your intended style more accurately.
Why Captions Matter (and Why You Should Review Them)
When you upload your images, Layer will automatically generate captions to describe the content and visual characteristics of each one. These help guide the AI during training.
For backgrounds, it’s especially important to review and edit these captions. Here’s what to check:
Is the perspective clearly stated? If your backgrounds are isometric, top-down, or side-view, make sure that’s mentioned. Inconsistent angles in the dataset or captions can lead to unpredictable results.
Does it capture unique visual traits? If your style includes things like a colored light source, exaggerated shadows, or a stylized texture, mention it in every image caption. You can also include this detail later as a Prompt Suffix to reinforce it during generation.
How to Start Training
Once your images are captioned, you’re ready to begin training.
Follow the general steps outlined in our How to Create a Custom Style article:
 | Go to the Art Styles page and click Build a New Style |
 | Upload your assets or create a style from selected files in your Layer Drive |
 | Choose a Style Base (SDXL, BRIA, or FLUX) |
 | Select Backgrounds as your Style Type |
 | Add at least 3 example prompts to show what you want to generate. |
 | Click Train - and you’re done |
You’ll get notified when the style is ready to use in Forge.
Best Practices for Background Style Training
 | Train with images that match your generation goals |
 | Keep perspective consistent |
 | Call out the art style |
 | Avoid mixing drastically different styles |
While It’s Training: Set Up Prompt Prefix + Suffix
If your captions share a consistent structure (like always starting with “An isometric view of…”), it’s a good idea to turn that into a Prompt Prefix.
Likewise, if your backgrounds all share an art direction - like “stylized 2D art with warm lighting and thick outlines” - use that as your Prompt Suffix.
This helps reinforce consistency when prompting later.