Adobe Firefly

Adobe Firefly is a family of generative artificial intelligence models developed by Adobe Inc., the American software company known for creative applications such as Photoshop and Illustrator. Firefly generates and edits images, illustrations, vector graphics, text effects, audio and video, and is embedded across Adobe's creative products as well as offered through a standalone application and commercial APIs. It is distinguished from many competing generative systems by Adobe's focus on what the company calls commercially safe AI: training primarily on Adobe Stock, openly licensed and public-domain content, and attaching provenance metadata to outputs.^[1]

History

Adobe announced Firefly in March 2023 as its response to the rapid rise of generative image tools such as Midjourney and Stable Diffusion. The first Firefly Image model reached general availability later in 2023, and Adobe quickly integrated generative features into its flagship tools — most visibly Generative Fill in Photoshop, which lets users add, remove or extend elements of a photograph with text prompts. At its MAX conference in October 2023 Adobe introduced Firefly Image 2 with improved photorealism, and in 2024 released Firefly Image 3 alongside a range of new product integrations. The company previewed a Firefly video model in late 2024 and commercialised video-generation features through 2025, adding capabilities such as extending video clips and generating b-roll from prompts inside Premiere Pro. In 2025 Adobe also launched Firefly as a standalone application and expanded Firefly Services, an API offering that lets enterprises build generative workflows into their own systems.^[1][2]

Key Concepts and Technology

Firefly's defining concept is the commercially safe model: Adobe trained initial Firefly models only on content it had rights to use — its licensed Adobe Stock library, openly licensed works and public-domain material — rather than indiscriminately on images scraped from the web, a practice that generated copyright litigation and controversy for several competitors. Adobe later added a mechanism for artists to opt in or out of having their work used in training, with contributors to Adobe Stock compensated when their content influences generated results.

Every Firefly output carries Content Credentials, a tamper-evident provenance label built on the C2PA open standard, recording that an asset was generated or edited by AI. Adobe has been a leading corporate advocate for content provenance, arguing that transparency about AI generation protects creators and helps platforms and publishers distinguish authentic work from synthetic content. Firefly models also produce vector graphics — native, scalable illustrations for logos and layouts — in addition to the raster images, text effects and video that other generative systems offer, reflecting the needs of Adobe's professional design customer base.^[1][3]

Applications and Impact

Firefly is used across the professional creative workflow. Photographers and retouchers use Generative Fill for cleanup and composition changes in Photoshop; designers generate layout imagery, illustrations and mock-ups in Express and Illustrator; marketers produce campaign variations at scale through Firefly Services; and video editors generate or extend footage in Premiere Pro. For enterprises, Adobe offers indemnification under its Firefly for Enterprise terms, promising legal protection if generated content leads to third-party intellectual-property claims — a decisive factor for brands and agencies that cannot risk copyright exposure. The main criticisms of Firefly centre on quality and stylistic control compared with specialist models such as Midjourney, and on broader industry concerns that generative tools reduce demand for stock photography and commissioned illustration.^[2][3]

>See Also

• Generative AI
• DALL-E
• Midjourney
• Stable Diffusion
• AI Watermarking