On March 21, 2023, Adobe unveiled Firefly, a family of generative AI models for creative work, and tied its pitch to a single differentiator: legal safety. While rivals like Stable Diffusion and Midjourney were trained on images scraped broadly from the web - and were already drawing copyright lawsuits from artists - Adobe said its first Firefly model was trained on Adobe Stock images, openly licensed content, and public-domain content whose copyright had expired. Adobe positioned the output as safe for commercial use.
The first model focused on generating images and text effects, with plans to fold Firefly-powered features into Creative Cloud, Document Cloud, Express, and Experience Cloud. The most visible product result was Generative Fill, which brought prompt-based, context-aware editing directly into Photoshop later in 2023, letting users add, remove, or extend parts of an image by describing them. Adobe also pushed Content Credentials, a provenance tag attached to generated and edited images, and advocated for a universal “Do Not Train” marker so creators could opt their work out of model training.
Firefly mattered less as a technical leap than as a positioning move. By controlling its training data and indemnifying enterprise customers against copyright claims, Adobe aimed Firefly squarely at professional and corporate users who could not risk using a model of uncertain provenance.
Why business readers should care: Firefly turned training-data provenance into a product feature and a sales argument. For commercial buyers, where an AI model’s training images came from - and who might sue over them - can matter as much as the quality of what it generates.