Get help faster and easier

New user?

Last updated on Sep 11, 2024

Learn to train custom models on the colors, shapes, and background aesthetic and generate images with the trained style using text descriptions.

Try it in the app
Create a custom model with images of your brand's assets in a few simple steps.

Before you begin:

Determine what style you want to train your model on and collect your assets. Cropping images to prepare them for training allows the model to focus on important visual elements.

  1. Go to the Firefly website and select Custom models from the top menu bar.

  2. On the Your models page, select Train a model and then give a name and description for the model.

  3. Select a Training mode. Select Style to train the model on the colors, shapes, and background aesthetic, or Subject to train the model on an object or character. 

    Create a custom model prompt is open and there are different fields like name, description, and training mode. Currently, the Training mode selected is Style.
    Select Style as the Training mode to train the model on the colors, shapes, and background aesthetic.

  4. In the Concept field, type a keyword for the main style. For example, if your model is to be trained on your product, type Style1.

  5. Include a sample prompt that will serve as a starting point for users to generate images using it.

    Give a sample prompt to show the custom model’s practical use and help users start generating images.
    Add a sample prompt to show the custom model’s practical use and help users start generating images.

  6. Drag and drop at least 10-30 images to train the model. Explore best practices for training efficient Firefly custom models.

    A Custom model called Food Campaign Poster is open and there are ten different photographs of various dishes, including pasta, salads, soups, and other meals uploaded.
    Include 10-30 images with substantial variation in context and subject matter while maintaining a consistent aesthetic.

  7. Once the images are uploaded, they are auto-captioned. You can select the image to view it and edit the caption to accurately describe its content, including key elements and actions. 

    One of the uploaded images had its caption updated for better accuracy. The UI shows that the updated caption is currently being applied.
    Uploaded images are automatically captioned, but you can use the Captions field to edit and enhance the description of the image.

    Tip:

    Custom models trained on Firefly Image 3 Model perform best with captions that are simplified and contain fewer than 15-18 words.

  8. Select Train, and Firefly will start the training. Once the training is complete, it will reflect on the Your models page with a Ready status.

    The UI shows “Training in progress.” Below it, there’s a message stating “Food Campaign Poster is training, you can leave this page.”
    While the training is in progress, you can leave this page or return to the model inventory page.

More like this

Get help faster and easier

New user?