Adobe Firefly: AI Personalization with Your Own Creations in Beta

Adobe recently announced the public beta release of its custom Firefly models, a significant advancement for its AI image generator. This feature allows creators and brands to train the model on their own works, ensuring that the generated images maintain a consistent aesthetic for characters, illustrations, and photography.

The primary goal of this tool is to streamline workflows for creative teams that need to produce a large volume of content. By offering a reusable foundation, custom models help preserve visual consistency across multiple projects, avoiding the need to start from scratch each time. Adobe emphasizes that these models can maintain essential details such as stroke weight, color palettes, lighting, and character features across image generations.

A crucial aspect of this new feature is privacy. Custom models are private by default, meaning that the images used for their training will not be incorporated into Adobe's general Firefly models. This approach aims to provide an ethical and commercially safe alternative compared to other services that might use copyrighted works without permission.

In its press release, Adobe stated: "To build a brand, you need a constant flow of assets that consistently express who you are. These assets should belong to you and you alone." Once a model is trained, it becomes an integral part of the creator's workflow, allowing for the generation of new ideas aligned with the desired aesthetic, reusing the model across various projects, briefs, and campaigns, and producing at scale without losing what makes the work distinctive.

The custom Firefly models were initially announced as a private beta during Adobe Max last year. Today, this feature is accessible to everyone. Adobe has long promoted its Firefly models, which are trained using a mix of licensed content and public domain works, as an ethical and secure solution.

While giving creative professionals more control over how models are trained seems like a natural extension, Adobe does not specify whether this will prevent users from training custom models on works they do not own. According to Adobe's help page, users will need to confirm that they have the necessary rights and permissions before training a custom model, and that their use will not infringe on the copyright, intellectual property, image, or privacy rights of others. We reached out to Adobe to inquire whether there are measures in place to prevent the training of custom models on a creator's work without permission.