On the Adobe Firefly homepage, select Custom Models.
Last updated on
Mar 19, 2026
Learn about custom models and how to use them for better control when generating variations that reflect your unique illustration style, subject, or characters.
With Firefly Custom Models, you can generate image variations that consistently reflect your unique illustration style, subject, or characters when using the Text to image feature in Firefly and Firefly Boards. Custom models help you explore new ideas, visualize different environments, and generate consistent content that is personalized to how you train them.
Note
Custom models are currently available in beta for Creative Cloud Individual or Creative Cloud Pro & ProPlus for Teams subscribers. Enterprise customers can learn more about and buy custom models today.
Training custom models involves the following process:
Browse the use cases for which you can train models. Select one of the use cases:
Illustration style
Train on your art style to create fresh designs that carry your signature look.
Photographic style
Train on your photos with unique lighting, color, and composition to capture your visual tone.
Character
Train on a specific character and bring them to life consistently across scenes and stories.
Drag and drop 10-30 images to train the model, following best practices for training custom models.
Make sure the images you add meet the following criteria:
- Aspect Ratio: Maximum of 16:9
- File Format: JPG or PNG
- Resolution: Minimum of 1000 pixels
You may encounter errors if the images contain issues such as low resolution. These should be corrected before beginning the training.
You'll see a training set score to the left of the train button. You can select it to open a panel that provides specific suggestions for modifying your training data to improve your model's ability to produce high-quality results. It scores the training set as a whole, assessing it against best practices for training custom models. It will provide you with an overall score and specific recommendations on how to improve it.
For example, avoid training a model on a spec-specific product, remove images that don’t align with the style of the other images, etc. The best practice is to use training sets that achieve a score of 85 or higher.
Based on what you upload, we will automatically generate the following information for you to review and edit:
|
Name |
Function |
|---|---|
|
Model title |
The title you will see when selecting your model. |
|
Model description |
A description of what the model generates. |
|
Model concept |
The name you'll use to refer to the specific subject (i.e., person or character) you’ve trained a model on. You’ll use this in your prompts, so choose something unique. This only applies to models trained on subjects, not styles. |
|
Sample prompt |
This gives individuals who use your model a starting point to decide what prompts to use with your model. |
|
Model tags |
Permanent attributes of the subject or style you're training a model on, such as brown hair for a brunette character. Do not include changeable attributes in Tags, like what object a character is holding. |
|
Captions |
Notable parts of each training image are described using language similar to what you will prompt with, such as “colorful illustration of a dog wagging its tail.” Models trained on a character should include the Model Concept in each caption. |
- The model title, model description, and sample prompt are not included in the actual training of the model and do not impact what it generates.
- If you edit captions and tags, select Analyze to reanalyze your changes before training.
Once you’ve uploaded your images and reviewed the information, such as model tags and captions, you can select Train from the top right.
The training may take anywhere from 30 minutes to a couple of hours, depending on the complexity of your model and the number of models under training.
You'll see a progress indicator that estimates the amount of training you have completed. You may close this tab while your model trains and then reopen it from Model Inventory. You'll also receive a notification to your email once your model has finished training.
Use the following model statuses to understand the progress and current state of model training:
- Draft: Status before training.
- Trained: Once the model is trained successfully.
- Cancelled: It may be marked as cancelled if you choose to cancel its training or if it encounters an unexpected error that requires attention before resubmitting it for training.
Note
- Training a model consumes generative credits. Learn about what generative credits are and how to use them.
- Generative credits are consumed as soon as you start training a model. If you cancel a model training, you'll not receive a refund for the credits. If model training fails due to an internal error, you'll be refunded the credits.
- Before training a custom model, you’ll be prompted to confirm that you have the necessary rights and permissions for the images you upload, that images of people may be analysed to create the model, and that your use of custom models won’t infringe on the copyright, IP, likeness, or privacy rights of others.
Once a model has finished training, you can use the model to generate images from a few places:
Select your new model from Model Inventory.
Select Create, Generate, or Boards from the left panel in Firefly to then select the custom model you've created.
Follow the required steps to generate images with your custom model. You will get two images per generation.
After generating the images, you can repeat as desired, or:
- Download by selecting the download icon .
- Open images in Express or Photoshop on the web to add finishing touches.
Tip
- Generating images consumes generative credits. Learn about what generative credits are and how to use them.
- When generating images with a model trained on a character, the model will remind you to include the model Concept in the prompt.
- If desired, you can adjust settings such as the Aspect ratio, Content type, Composition , Style references, and Effects as you generate.
You can choose to share your models with others to use to generate images in Firefly or Firefly Boards.
On the Your models page, identify a custom model with a Trained status and select Share from the More menu.
On the Invite people dialog box, under the Can use tab, enter one or more email addresses of the users who will be granted access to use the custom model.
If needed, add a note for the invited user, then select Invite to generate.
Note
Sharing a model with another person to generate images does not grant them access to edit your model or share it with others.
When you train a custom model, it's saved to a project to keep your work organized. You can access and manage your projects and the custom models in them by visiting Adobe Home and navigating to Files > Projects. You can view all your projects, as well as those to which you have been invited to collaborate.
Users with access to the project where a custom model is stored will automatically inherit access to that model and any other models saved within it. You can move your model to a new folder, if desired.
You can duplicate or delete a custom model. To do so, select a custom model, go to the More menu, and choose one of these options:
- Duplicate: Copy the images used from a model to start a new model for training.
- Delete: Delete and remove the model with the assets and captions you used to train it.
Note
You cannot retrain a trained custom model.
Turn text into art with Adobe Firefly
Create beautiful images and video clips from text prompts using generative AI.
