Adobe has launched Firefly Custom Models in public beta, allowing creators and businesses to train AI image generators using their own visual assets. The tool lets users teach the AI to recognize and reproduce specific styles, character designs, color palettes, lighting conditions, and stroke weights.
The goal is to help teams produce large volumes of content that stays visually consistent across campaigns, platforms, and markets. Once trained, a custom model can be reused across multiple projects without starting over each time.
Custom models are private by default. Adobe states it will not use assets submitted for model training to improve its general Firefly AI models. The models require an enterprise license and integrate with several Adobe tools, including Adobe Express, GenStudio for Performance Marketing, and Firefly Services.
Training requires high-quality JPG or PNG images with a minimum resolution of 1024×1024 pixels and files under 50 MB.
On the question of misuse, Adobe asks users to confirm before training that they hold the necessary rights to the images they upload. However, Adobe has not disclosed what technical measures, if any, are in place to prevent users from training models on artwork they do not own.
Sources: The Verge, Adobe (official product page)
