Hi, how can we help you today?
Hi, how can we help you today?
A Style Model is a custom-trained AI model designed to replicate a specific aesthetic, whether it’s a color palette, brushwork, outlines, cartoon style, 3D rendered, realistic, or more. By learning from a curated dataset with consistent stylistic traits, it generates content that maintains the same visual identity.
The same prompt is applied to a custom-trained model (left) and to a base model (right). The custom-trained model immediately reproduces the style of the training images, even with a short prompt
Style Models have a wide range of applications: they can create uniform backgrounds and environments, generate series of multiple characters with matching proportions and style, design assets for games or animations, produce consistent illustrations for storytelling, or ensure a cohesive look across different scenes and elements in a project.
Some examples of Style Models available on Scenario
These models offer powerful control. Instead of relying on unpredictable style prompts, you can focus mostly on the subjects or scene composition, knowing the AI has already "learned" the style and should naturally apply it to every generated image.
Style Models are also the easiest type of custom AI models to train, but following some best practices is key. The most critical factor is curating a high-quality, diverse dataset. The success of your model depends largely on selecting the right images and adjusting training parameters.
A Style Model is essential for maintaining a pre-existing aesthetic from a game or brand, especially when that style isn’t achievable through prompting foundation (base) models.
Style-consistent images generated with the “3D Blocky Elements” model on Scenario
This guide walks you through the steps to create high-quality Style Models effectively.
A well-curated dataset is the foundation of a great Style Model, and you should follow the following rules:
Image Quality:
Use high-resolution images (1024 pixels or higher) to ensure the model captures fine details like textures and brushstrokes. If your images are too small, you can use the Enhance tool to upscale them.
Consistency in Style:
All images should share a cohesive aesthetic, whether through color palette, lighting, or artistic techniques. This helps the model apply the style consistently across different contexts.
Variety in Context:
While keeping a consistent style, include diverse subjects, environments, and perspectives. A strong dataset features different objects, scenes, and angles within the same style, making the model more versatile. Avoid excessive repetition (such as the exact same depth or composition), as it can limit the adaptability of the model.
In this example , the dataset to the left is made with all very consistent images (same style, same proportion, same angle of view). The one to the right has images that do not share the same style
When deciding on the size of your dataset, “less is more”. A small, well-curated set (10-20 images) will usually yield better results than a larger dataset (30-100 images) that lacks enough variety or contains too many similar examples.
For beginners, we recommend starting with 10 to 15 high-quality images. As you gain experience, you can gradually expand your dataset to enhance the model’s capabilities. If your images are both consistent and diverse, it's better to stay on the smaller side (20 max) rather than using an overly large set.
Even just 10 images can give you a great style model, like this training dataset for “Top-down TD Game” on Scenario (link)
All training images must be square. You have two options: (i) crop your images to a square format before uploading or (ii) upload images in any format and adjust cropping directly in Scenario’s interface during the upload process.
If you upload landscape or portrait images without adjusting the crop, the entire image will be fitted into a square, by default. For greater flexibility (especially when training a model on a specific character style) you can mix image formats, using both “square“ and “landscape“ ratios where needed, as shown in this example below where various poses are available (close-up portrait, full body views, etc):
Once your dataset is uploaded, caption your images. Captions provide context, helping the AI understand key stylistic elements such as structure, lighting, and colors.
Scenario offers an instant, automated captioning tool, which works well in the vast majority of cases. However, reviewing automated captions is recommended. You can edit them by clicking on each image individually. For more details on best captioning practices, refer to our dedicated guide.
Select a base model for training. You can choose from Flux, SDXL, or Bria (if activated in your account). Each base model has different characteristics and you might even want to train your custom model on both and compare results:
Flux (default) – The most up-to-date model, highly versatile, and working with a large range of captions. It offers great prompt adherence and allows for longer prompts while maintaining strong consistency. However, it might often “lean toward realism” unless style elements are explicitly prompted.
SDXL – Less versatile and with a lesser prompt adherence, but better at maintaining the simple, “minimalist” styles. It works best with carefully tailored captions, especially for character styles (see guide).
Bria – Reach out for more information (Enterprise users).
Once everything is set, click "Start Training" and wait for the process to complete. The training time will depend on your dataset size, number of training steps, and base model. You’ll be notified via email and through the “Recent Tasks“ (three-bar) icon to the top menu, which will display a red dot when training is complete.
Don’t forget to finalize the process by testing your model, making refinements if needed, and adding a description, tags, and pinned images. Please see this guide for more information on how to properly refine and manage models.
You can follow our API recipe to train a model: https://docs.scenario.com/recipes/train-a-flux-dev-lora-model
Resources:
Create a Model: Scenario API Documentation - Create a Model
Add Training Images: Scenario API Documentation - Add Training Images
Modify Captions: Scenario API Documentation - Modify Captions
Trigger Model Training: Scenario API Documentation - Trigger Model Training
Was this helpful?
Quentin