Adobe Launches Firefly Custom Models Public Beta

Adobe has launched a public beta for Custom Models in its Firefly generative AI platform, enabling users to train artificial intelligence systems on their own images to produce visuals in specific styles for illustration, photography, and character design. The update, announced this week, aims to help creators and brands maintain consistency across projects while integrating new tools and partnerships to expand Firefly's capabilities.

Custom Models allow individuals and enterprises to upload a set of reference images, typically ranging from 10 to several thousand, to fine-tune Firefly's image generation. Users provide JPG or PNG files with a minimum resolution of 1024x1024 pixels and consistent aspect ratios, focusing on elements like color palettes, lighting, and compositional styles. Adobe states that most models train in under an hour, generating outputs that preserve distinctive features such as stroke weights or product details. For example, a fashion brand could create images of a specific sneaker design in various settings without manual recreation. The feature is available to premium subscribers, with each training session consuming 500 generative credits from a user's...